# AgentDrives > Make sure AI agents can find your product, run it and pay you. Our agent signs up for each product, does real work, tries to pay and leaves, and we publish how far it gets on its own, with the evidence. Agent levels: L0 Read-only (Agents can read about it, but there is nothing to use.); L1 Needs a person (Agents can't get in on their own.); L2 Gets in (An agent gets in on its own but does less than half the core work.); L3 Does the work (An agent gets in on its own and does half the core work or more.); L4 Runs on its own (An agent does the core work and can stop paying on its own.); L5 Autonomous (All of that, and it pays on its own for the work.). Part scores, 0-100, next to the level (never a total, never in the level; null = not measured): Holds up (Reliability and the crash test); Docs match (The docs checked against live behaviour); Agent-ready setup (What an agent finds to start with). Security: reported privately to the owner, the same for every product. Data: https://agentdrives.com/api/tools.json?via=llms (CC BY 4.0). Every tool page has a markdown twin at /tools/.md. ## For agents Sign up in one call, with no person and no address: POST https://agentdrives.com/api/agents/signup {"agent":"your-name"} returns an API key at once and a claim link for your person (MCP tool: sign_up). Until a person claims the account or it pays for a Drive, it has 1 free check a day and a Drive paid over HTTP 402. Everything else works without a browser too: sign in with an emailed code, create an API key, start a free check of any product, claim a domain, prove it is yours (an account email on the domain, which verifies at once, a code mailed to any address at the domain, a DNS TXT record, a meta tag on the home page or a file at /.well-known/agentdrive-verify.txt), order drives, cancel, delete your account with your key (paid drives not run yet are cancelled and refunded; GET /api/me/delete lists them first). A paid order before the domain is proved gets 403 claim_not_verified with the fix. Guide: https://agentdrives.com/data?via=llms#agents · OpenAPI: https://agentdrives.com/openapi.json?via=llms · MCP: https://agentdrives.com/mcp (send x-api-key for account tools); the directory tools alone, without a key: https://agentdrives.com/api/mcp. Pay without a person: a Drive order returns payment.agent. POST https://agentdrives.com/api/orders/{id}/payment answers 402 for $49 with MPP challenges (WWW-Authenticate: a card through a Stripe Shared Payment Token, or USDC on Tempo or Base) and x402 (PAYMENT-REQUIRED: USDC on Base); send the request again with the credential and the drive is queued (MCP tool: pay_order). The 402's fix names the rails that are on. The price includes any VAT. For our invoice, send {"buyer": {"vat_id": "DE123456789", "company": "Acme GmbH", "address": "1 Main St, 10115 Berlin"}} as the JSON body of both requests, or with the order: an EU VAT ID that VIES confirms gets it at once, with no VAT outside the Netherlands. Every payment has a receipt at https://agentdrives.com/api/orders/{id}/receipt. Pay per call: $0.01 a search, no rate limit, from prepaid credit or one call at a time. Credit: POST https://agentdrives.com/api/credits/topup {"amount_usd": 5} answers 402 with MPP (a card through a Stripe Shared Payment Token, or USDC on Base or Tempo) and x402 (USDC on Base); pay it and send the request again with the credential. Send an Idempotency-Key header with both requests: a retry never pays twice. $5 buys 500 calls, $20 buys 2,000, $50 buys 5,000. A balance holds at most $100 and lasts 12 months from the last top-up. It pays for calls only, not a Drive. The price includes any VAT, and every top-up has a receipt. Without a key the top-up makes an agent account and returns its key (MCP tool: top_up). Then POST https://agentdrives.com/api/paid/find-tools {"task": "send an invoice"} with x-api-key answers at once and says what's left (credit in the body, X-Balance-Remaining); GET https://agentdrives.com/api/credits shows the balance (MCP tool: balance). One call at a time instead: without a key the call answers 402 with x402 and MPP, USDC only (a card's minimum is above the price). The same search is free at GET https://agentdrives.com/api/find-tools, rate-limited per IP. ## Pricing Start free. Then let a real agent drive it. Find out where AI agents stop in your product, and what to fix first. One price per product. The whole offer as JSON: https://agentdrives.com/pricing.json?via=llms. Order over the API (POST /api/orders, product ids below), the MCP order tool, or a person in the dashboard: https://agentdrives.com/app/drives?plan=. - Free check ("product": "free"): $0, private. See what an agent sees. What an agent finds without a browser: docs, API, MCP, llms.txt; The wall it would likely hit, and the first fixes; How the products we drove in your category did; Whether your product can be listed. In minutes. One per domain every 90 days. In the directory: nothing, it stays private. - Drive ("product": "drive"): $49 a drive. Where a real agent stops in your product, and what to fix first. Where agents stop, with the evidence, and the three fixes that move your level; Every way in tried: email, Google, GitHub, your API, CLI and MCP; Your core tasks done and read back; Your checkout walked to the last click, and a paid call where you sell one; Up to 600 pages of your docs read and judged; Cancel, delete, the crash test, and which agents get past your bot check. Your level and a dated badge. Listed, or private: your choice. In the directory: listed with your level. - Audit ("product": "audit"): $299 one time. How agents do with your whole API, MCP server and CLI, next to your rivals. Everything in a Drive; Every documented API route, MCP tool and CLI command checked against your docs; A security review, for you only; Up to three rivals, compared privately; A second drive within 30 days to prove the fix. For when your API, MCP server or CLI is the product. In the directory: listed with every part score. - The free check needs no domain proof: any signed-in account (offers_ok: true only if you want the monthly report and offers); one per domain every 90 days (within them the call answers with the domain's latest check and next_free_at) and 3 a day per account. A Drive or an Audit needs a verified claim on the product's domain (403 claim_not_verified otherwise); an Audit needs security_ok: true (the go-ahead for the security review) and takes rivals: [url or slug, up to 3] and the owner's paid plan at cost (plan). - Monthly drive ("product": "monthly", or "yearly" for a year): $29 a month or $290 a year, per product. Let us drive it every month. Our agent drives your product's path every month; A message when your level or the place an agent stops changes; Driven monthly on your listing and badge. For a product in the directory that your account has claimed and verified (409 not_listed otherwise); one per product; 10 or more products: write to hello@agentdrives.com. Cancel any time; the listing stays. - Again: Drive it again any time for $49, or $29 a month and we drive it every month. - Upgrade: Within 30 days of a Drive's report, an Audit costs $250: what you paid counts. Order the Audit with upgrade_from: the Drive's order id (GET /api/me lists each order's upgrades and their prices). Once per order. - Re-drives: A listing never lapses because nobody pays. It shows the date of its drive. Fixed something? Order a Drive ($49) any time, or the Monthly drive ($29/month). - How the directory works: A product is in the directory because an agent drove it. We pick products ourselves, and anyone can suggest one for free; an owner can also order a drive of their own product. Both drives are the same, listed at whatever level they find, and an owner's order is labelled. An owner's order says "Owner-paid drive" on its listing and in the API. Suggest a product: POST /api/suggest {url, email?}. - Stack check · 3 tools: $490 one time. Three of your candidate tools, driven side by side; Which ones your agents can sign up for, get a key and do the core work in; Where an agent stops in each, and which kind of agent gets furthest. - Stack check: $1,490 5 tools · +$250 per extra. Five of your candidate tools (up to ten), driven side by side; Which ones your agents can sign up for, get a key and do the core work in; What each one needs from you to work for your agents; A walk-through of the result. - Fair use (Audit): One Audit covers one product: one sign-up, and the API, MCP server and CLI that come with it. We read and check every documented operation, whatever the size, and run every one that is safe on our test account within one drive of up to three hours. On very large products we run all reads, everything behind your core tasks and at least one call per resource group, and the report says exactly what was sampled. Your second drive picks up the rest. - Badges: Driven (a drive, with its level and date) and Driven monthly (while a Monthly drive runs). - Not an owner: Stack check (above) for teams choosing tools for their agents; data plans for agent builders at https://agentdrives.com/data?via=llms. - Crash test (the resilience part): every drive tests how the product handles an agent's mistakes once it is in, on our own test account: a few malformed requests, a wrong key of its own, a retried create, 10 to 20 reads of one endpoint (stopping at the first 429) and each core feature run three times. - Prices are in US dollars, before VAT: 21% Dutch VAT in the Netherlands, and elsewhere in the EU without a VAT ID; none for a business elsewhere in the EU with a VAT ID (reverse charge) or a buyer outside the EU. Every card checkout comes with our invoice; paid by an agent over HTTP 402, the price includes any VAT. The report within 24 hours of payment, or your money back. Details: https://agentdrives.com/pricing?via=llms ## Tools - [Parallel](https://agentdrives.com/tools/parallel.md?via=llms): L5 · Autonomous, Search APIs. An agent can search and extract through Parallel's free anonymous MCP server and pay per call with x402 or MPP for search, extract, responses and research, all without an account. - [Attio](https://agentdrives.com/tools/attio.md?via=llms): L4 · Runs on its own, CRM. Once a token exists, an agent can run almost all of Attio's CRM core over a clean, well-documented REST API and an official 41-tool MCP server: every core capability was done and verified. - [Steel](https://agentdrives.com/tools/steel.md?via=llms): L4 · Runs on its own, Browser infrastructure. An agent can do almost all of Steel through its API, SDK, CLI or official MCP server: sessions over CDP, scrape/screenshot/PDF, profiles, stored logins, files and extensions all worked and read back. - [AgentMail](https://agentdrives.com/tools/agentmail.md?via=llms): L4 · Runs on its own, Email infrastructure. An agent can sign up, verify, get a key and run all nine core email tasks headless through the API, the official MCP server (the 28 Jul 2026 spec) and the CLI, with clear errors and working idempotency. - [Cloudinary](https://agentdrives.com/tools/cloudinary.md?via=llms): L4 · Runs on its own, File storage. One unauthenticated API call gives an agent a working Cloudinary cloud, and from there all 9 core capabilities ran headless through REST, the CLI and three remote MCP servers, verified by reading state back. - [Airtable](https://agentdrives.com/tools/airtable.md?via=llms): L4 · Runs on its own, Databases. An agent can run almost all of Airtable's core once it has a token: REST and the official MCP server (46 tools) covered bases, records, schema, webhooks, automations, interfaces and forms, and Omni and field agents worked in the browser. - [Basecamp](https://agentdrives.com/tools/basecamp.md?via=llms): L4 · Runs on its own, Project management. Basecamp is one of the most agent-ready products we've driven: a plain form POST signs up a free account, and the official CLI with its built-in MCP server, the REST API and the SDKs covered all nine core capabilities. - [DocuSeal](https://agentdrives.com/tools/docuseal.md?via=llms): L4 · Runs on its own, E-signature. An agent signed up, got an API key (after an email code) and ran the full template-to-signed-PDF workflow headless through the documented REST API; all 7 core capabilities verified, with signer steps done in a browser by design. - [Zulip](https://agentdrives.com/tools/zulip.md?via=llms): L4 · Runs on its own, Team chat. Zulip Cloud Free can be run almost entirely by an agent through its documented REST API: all 7 core capabilities and most secondary ones were done and read back, and the organization was deactivated by API. - [Formbricks](https://agentdrives.com/tools/formbricks.md?via=llms): L4 · Runs on its own, Forms. An agent can run Formbricks nearly end to end. - [NocoDB](https://agentdrives.com/tools/nocodb.md?via=llms): L4 · Runs on its own, Databases. Once it holds a token, an agent can run most of NocoDB on the free plan: bases, tables, fields, records, views, sharing, CSV import, workflows, interfaces and dashboards all worked, mostly over the official MCP server, and NocoAI built a table from one sentence in the web app. - [Firecrawl](https://agentdrives.com/tools/firecrawl.md?via=llms): L4 · Runs on its own, Web scraping. Firecrawl's core works for an agent: scrape, crawl, map, search, extract, actions, agent and parse all ran and read back on the free plan, and scrape/search/parse even work keyless through MCP or the API. - [Pipedream](https://agentdrives.com/tools/pipedream.md?via=llms): L4 · Runs on its own, Automation. Pipedream Connect's core (connected accounts, tool runs via REST and MCP, proxy, triggers, webhooks) worked headless and verified on the free plan once credentials existed. - [Browserless](https://agentdrives.com/tools/browserless.md?via=llms): L4 · Runs on its own, Browser infrastructure. With a token, an agent can do nearly everything in Browserless headless: REST, BrowserQL, Puppeteer, the Agent Run API and a 16-tool MCP server all worked on the Free plan and were verified. - [Tally](https://agentdrives.com/tools/tally.md?via=llms): L4 · Runs on its own, Forms. Once a browser sign-up has produced an API key, Tally is very easy for an agent to work with. - [Anchor Browser](https://agentdrives.com/tools/anchorbrowser.md?via=llms): L4 · Runs on its own, Browser infrastructure. Anchor Browser lets an agent in with no account at all: a short text puzzle over HTTP returns an API key, and from there sessions, CDP control, natural-language tasks, saved tasks, managed logins, profiles, files, webhooks and batches all worked headless on the free tier (9 of 10 core capabilities verified). - [Plane](https://agentdrives.com/tools/plane.md?via=llms): L4 · Runs on its own, Project management. Plane is highly agent-operable once a key exists: every core task except AI agents ran end to end over the documented REST API (and the official MCP server), three times each, with clean read-back. - [GitHub](https://agentdrives.com/tools/github.md?via=llms): L4 · Runs on its own, Project management. An agent cannot open a GitHub account: the sign-up page answers our signed agent with a DataDome block. - [Shortcut](https://agentdrives.com/tools/shortcut.md?via=llms): L4 · Runs on its own, Project management. Once the agent had signed up and created an API token in its browser, Shortcut is very agent-friendly: an accurate v4 OpenAPI, clear JSON errors and an official MCP server with dynamic OAuth registration let us create, move, link and clean up stories, epics, iterations, objectives and docs headless. - [Buttondown](https://agentdrives.com/tools/buttondown.md?via=llms): L4 · Runs on its own, Email marketing. Buttondown is very agent-friendly once the key is in hand: a complete OpenAPI, clear errors that name the fix, and on the free plan we drafted, scheduled, sent, personalised, archived, exported, imported and bulk-edited through the API, all verified by reading back. - [Fillout](https://agentdrives.com/tools/fillout.md?via=llms): L4 · Runs on its own, Forms. An agent can sign up and get a key in a browser, then handle submissions headless (create, list, search, delete, webhooks) through a small, accurate REST API. - [Twenty](https://agentdrives.com/tools/twenty.md?via=llms): L4 · Runs on its own, CRM. Once a workspace exists, Twenty is one of the most agent-ready CRMs: its official MCP server (324 tools) and REST/GraphQL let an agent run records, custom objects, views, workflows, AI-agent steps and dashboards end to end. - [SignWell](https://agentdrives.com/tools/signwell.md?via=llms): L3 · Does the work, E-signature. SignWell's documented REST API, official CLI and stdio MCP server let an agent do the whole core flow: draft, fields, signing order, send, track, signed PDF, templates and webhooks, free in test mode. - [Craft](https://agentdrives.com/tools/craft.md?via=llms): L3 · Does the work, Docs and workspace. Craft has a rich documented space API (44 operations) and an official OAuth MCP server. - [Plausible Analytics](https://agentdrives.com/tools/plausible.md?via=llms): L3 · Does the work, Web analytics. An agent can open a 30-day trial (no card) in a browser and then read its stats headless over the Stats API. - [Umami](https://agentdrives.com/tools/umami.md?via=llms): L3 · Does the work, Web analytics. An agent can sign up for Umami Cloud's free plan without a bot check and use the whole dashboard in a browser: websites, goals, funnels, notes, links, pixels, shares, export and account delete. - [Pipedrive](https://agentdrives.com/tools/pipedrive.md?via=llms): L3 · Does the work, CRM. Pipedrive is very drivable once an agent is in: through the free developer sandbox (one browser form, no CAPTCHA) and an API token, we created, moved, converted, searched, filtered and deleted CRM data end to end via REST v1/v2 and the official OAuth-protected MCP server. - [Coda](https://agentdrives.com/tools/coda.md?via=llms): L3 · Does the work, Docs and workspace. The agent signed up, verified its email and made an API token itself in a browser; from there it ran Coda (now Superhuman Docs) through the REST API and the official MCP server: a doc, pages, a typed table with formula and button columns, rows, sharing and a Pack, all read back. - [Convex](https://agentdrives.com/tools/convex.md?via=llms): L3 · Does the work, Databases. Convex is one of the most agent-ready backends to build on. - [Stripe](https://agentdrives.com/tools/stripe.md?via=llms): L3 · Does the work, Payments. An agent gets working Stripe test keys with one CLI command and no account, then drives payments, Checkout, Elements, subscriptions, invoices, coupons, refunds, webhooks and the official MCP server headless (6 of 9 core capabilities). - [monday.com](https://agentdrives.com/tools/monday.md?via=llms): L3 · Does the work, Project management. monday.com is unusually agent-ready once an agent is inside: llms.txt, agent skills, and a hosted MCP with 98 tools that accepts the API token directly. - [Mattermost](https://agentdrives.com/tools/mattermost.md?via=llms): L3 · Does the work, Team chat. An agent can sign up for a 30-day Mattermost Cloud Enterprise trial without a card, but only in the browser. - [Railway](https://agentdrives.com/tools/railway.md?via=llms): L3 · Does the work, Hosting and deploys. Railway is unusually agent-ready on paper (llms.txt, markdown docs, auth.md, API catalog, PRM, CIMD, device flow, official MCP), and an agent can get a free VM and a project token with no account via `ssh railway.new`, then add databases, set variables, open TCP proxies, read logs and metrics, and run sandboxes. - [TidyCal](https://agentdrives.com/tools/tidycal.md?via=llms): L3 · Does the work, Scheduling. On TidyCal's free plan an agent can do the core scheduling in a browser: create a booking type, publish it, and book, reschedule and cancel an appointment. - [TickTick](https://agentdrives.com/tools/ticktick.md?via=llms): L3 · Does the work, Task management. With a personal API key, an agent can drive TickTick's tasks, lists, tags, reminders, repeats, comments, Kanban, focus records and habit check-ins through the documented REST API or the official 28 Jul 2026 MCP server. - [Penpot](https://agentdrives.com/tools/penpot.md?via=llms): L3 · Does the work, Design. An agent can get through Penpot's whole access ladder headless with the documented RPC API: sign up, verify, token, first design, delete account. - [Tavily](https://agentdrives.com/tools/tavily.md?via=llms): L3 · Does the work, Search APIs. Tavily is easy for an agent to start with: keyless Search and Extract work on REST, a modern 28 Jul 2026 MCP server and the CLI with no account, and covered 4 of 8 core tasks. - [Mixpanel](https://agentdrives.com/tools/mixpanel.md?via=llms): L3 · Does the work, Product analytics. An agent can sign up for Mixpanel's free plan in a browser without a CAPTCHA, but the first credential (a service account) exists only in the web UI. - [Browse AI](https://agentdrives.com/tools/browse.md?via=llms): L3 · Does the work, Web scraping. Browse AI has a clean, documented REST API for running robots, monitors, bulk runs and webhooks; an agent drives all of that headless once a robot exists. - [Neon](https://agentdrives.com/tools/neon.md?via=llms): L2 · Gets in, Databases. Neon is unusually agent-friendly without an account: auth.md and `neon claim create` gave a working Postgres database, Managed Better Auth and a Data API in seconds, and we drove SQL, extensions, autoscaling, scale-to-zero, auth sign-ups, RLS-protected REST CRUD and neon.ts deploys headless. - [Brevo](https://agentdrives.com/tools/brevo.md?via=llms): L2 · Gets in, Email marketing. An agent can sign up for Brevo's free plan in a browser, verify by email headless, and get an MCP or REST token through OAuth with one browser consent, then manage contacts, lists, CRM deals and records, templates, webhooks and eCommerce data over a 282-tool MCP server and a 286-operation REST API. - [Shopify](https://agentdrives.com/tools/shopify.md?via=llms): L2 · Gets in, E-commerce. Without any account, an agent can create a temporary Shopify store from the CLI and run the Admin GraphQL API on it: products, variants, inventory, theme files, pages, metaobjects, bundles, shipping rates and discounts all worked and read back. - [Fly.io](https://agentdrives.com/tools/fly.md?via=llms): L2 · Gets in, Hosting and deploys. On Fly.io an agent can drive Sprites end to end over the documented REST API and the official Sprites MCP: create, exec, checkpoint/restore and a public URL. - [Zapier](https://agentdrives.com/tools/zapier.md?via=llms): L2 · Gets in, Automation. An agent can sign up and authorise Zapier in a browser (no CAPTCHA, no email confirmation needed) and then work headless through the SDK CLI and the official MCP server: we created, updated and read back records in Zapier Tables, ran a Tables action through MCP, and saved and used an MCP Skill. - [Exa](https://agentdrives.com/tools/exa.md?via=llms): L2 · Gets in, Search APIs. An agent can use Exa without an account: the hosted MCP server searches and reads pages with no key, and /search and /contents take x402 or MPP payments per call. - [Browserbase](https://agentdrives.com/tools/browserbase.md?via=llms): L2 · Gets in, Browser infrastructure. Browserbase sells browser sessions, search and fetch to agents with no account over x402 (v1) and MPP on its own domain; we paid one $0.01 session and the agent read it back and ended it. - [Gumroad](https://agentdrives.com/tools/gumroad.md?via=llms): L2 · Gets in, E-commerce. An agent gets a Gumroad account and a token with a few browser steps (sign-up form, token button), then works fully headless through a ~80-operation REST API, an official CLI and an official MCP server: drafts, variants, discounts, files, upsells, webhooks. - [Netlify](https://agentdrives.com/tools/netlify.md?via=llms): L2 · Gets in, Hosting and deploys. An agent can deploy a static site to Netlify with no account (netlify deploy --allow-anonymous) and use the Image CDN on it, and the docs are agent-ready (llms.txt, Markdown pages, OpenAPI, MCP server card). - [Vercel](https://agentdrives.com/tools/vercel.md?via=llms): L2 · Gets in, Hosting and deploys. Vercel's docs are very agent-ready (llms.txt, Markdown pages, OpenAPI, AI Catalog, official MCP with OAuth and a CLI device login), but an agent cannot get in: the email sign-up was refused with no reason, and the GitHub sign-up created the account and then asked for a phone number for an SMS code on every login. - [PlanetScale](https://agentdrives.com/tools/planetscale.md?via=llms): L2 · Gets in, Databases. An agent gets into PlanetScale entirely headless: sign-up form without CAPTCHA, email verification, CLI device login, service tokens, REST API and a 25-tool hosted MCP server, all with unusually agent-friendly JSON errors and an agent setup guide. - [Postiz](https://agentdrives.com/tools/postiz.md?via=llms): L2 · Gets in, Social media scheduling. Postiz is built for agents on paper (llms.txt, OpenAPI, CLI, official MCP with dynamic client registration), and sign-up plus email verification went through in the browser with no CAPTCHA. - [Webflow](https://agentdrives.com/tools/webflow.md?via=llms): L2 · Gets in, Website builders. Webflow has strong agent surfaces: Markdown docs with llms.txt, an OpenAPI 3.1 spec, and an official remote MCP server on the current 28 Jul 2026 protocol. - [Supabase](https://agentdrives.com/tools/supabase.md?via=llms): L1 · Needs a person, Databases. Once in, an agent can do nearly all of Supabase headless: all 10 core capabilities passed create-read-update-delete through the Management API, MCP (the 28 Jul 2026 spec) and CLI, with repeats passing except a third user signup that hit the built-in mailer's rate limit (GraphQL first needed pg_graphql enabled). - [Apify](https://agentdrives.com/tools/apify.md?via=llms): L1 · Needs a person, Web scraping. An agent with a wallet gets into Apify with no account and no human: we paid $1 over x402 at agi.apify.com and the token ran every core task headless (Actors, storage, own Actor build and deploy, proxy from an Actor), 8 of 9 core capabilities, each repeated three times. - [Intercom](https://agentdrives.com/tools/intercom.md?via=llms): L1 · Needs a person, Customer support. An agent can sign up for Intercom's no-card trial in a browser, take an API token from the Developer Hub web page, and then run the helpdesk headless: conversations, tickets, ticket types, contacts, articles and Fin's knowledge all worked by API and read back cleanly, and Fin and Copilot answered from content the agent added. - [Copper](https://agentdrives.com/tools/copper.md?via=llms): L1 · Needs a person, CRM. Copper's REST API is solid: once an agent has a key it can run the core CRM loop of contacts, deals, stages, won/lost, tasks, activities, custom fields and tags headlessly and verify every step. - [ActiveCampaign](https://agentdrives.com/tools/activecampaign.md?via=llms): L1 · Needs a person, Email marketing. ActiveCampaign's docs are agent-ready (llms.txt, markdown negotiation, OpenAPI 3.1.0, a docs MCP on the 28 Jul 2026 spec, auth.md, AI catalog), but an agent cannot get in: the free-trial sign-up hits a Cloudflare Turnstile challenge and credentials only come from the app UI. - [Acuity Scheduling](https://agentdrives.com/tools/acuityscheduling.md?via=llms): L1 · Needs a person, Scheduling. Acuity's docs are agent-friendly (llms.txt, Markdown pages, OpenAPI 3.1.0), but an agent cannot get in: the Squarespace sign-up form stalled silently behind an invisible reCAPTCHA Enterprise check, and robots.txt disallows /api/*, which covers the whole documented REST API. - [Akiflow](https://agentdrives.com/tools/akiflow.md?via=llms): L1 · Needs a person, Task management. An agent cannot get into Akiflow on its own: registration answers with a reCAPTCHA image puzzle, the page hides it and hangs. - [Amplitude](https://agentdrives.com/tools/amplitude.md?via=llms): L1 · Needs a person, Product analytics. Amplitude is unusually well documented for agents (llms.txt, Markdown pages, OpenAPI 3.1, an AI catalog, an official OAuth MCP server and a public Docs MCP), but an agent cannot get in: the only sign-up is a web form whose invisible reCAPTCHA rejected our agent's browser. - [Asana](https://agentdrives.com/tools/asana.md?via=llms): L1 · Needs a person, Project management. Asana has a large REST API, an official OAuth MCP server and agent-readable Markdown docs, but no way for an agent to get in: the sign-up form's reCAPTCHA Enterprise rejected our agent's browser and showed a visible checkbox. - [Backblaze B2](https://agentdrives.com/tools/backblaze.md?via=llms): L1 · Needs a person, File storage. An agent cannot open a Backblaze account: both sign-up pages show a reCAPTCHA v2 'Security Check' to our agent's browser, so we drove no capability. - [Baserow](https://agentdrives.com/tools/baserow.md?via=llms): L1 · Needs a person, Databases. Baserow has a large documented REST API (OpenAPI 3.0.3, 424 operations), an official MCP server and llms.txt, but an agent cannot create an account: sign-up, by API and by web form, needs a Cloudflare Turnstile token. - [beehiiv](https://agentdrives.com/tools/beehiiv.md?via=llms): L1 · Needs a person, Email marketing. beehiiv's docs are agent-friendly (llms.txt, Markdown twins, OpenAPI 3.1.0, docs and marketing MCPs) and it has an official OAuth MCP server. - [BigCommerce](https://agentdrives.com/tools/bigcommerce.md?via=llms): L1 · Needs a person, E-commerce. An agent cannot open a BigCommerce store on its own: the trial sign-up form shows a visible reCAPTCHA Enterprise checkbox, there is no headless sign-up, and the first API token is created only in the control panel. - [Box](https://agentdrives.com/tools/box.md?via=llms): L1 · Needs a person, File storage. An agent cannot get into Box on its own: both documented sign-ups (personal and free developer) put an hCaptcha checkbox in front of account creation. - [Brave Search API](https://agentdrives.com/tools/brave.md?via=llms): L1 · Needs a person, Search APIs. An agent can sign up and verify in a browser, but an API key needs an activated plan, and every plan, even the one with $5 of free monthly credits, needs a card at Stripe Checkout. - [Buffer](https://agentdrives.com/tools/buffer.md?via=llms): L1 · Needs a person, Social media scheduling. Buffer offers a GraphQL API, an official MCP server, a CLI with agent skill files and Markdown docs, but an agent cannot get an account: the web-only sign-up shows an hCaptcha picture puzzle. - [Cal.com](https://agentdrives.com/tools/cal.md?via=llms): L1 · Needs a person, Scheduling. Cal.com is very well prepared for agents on paper: AI catalog, MCP server card, auth.md, agent skills, llms.txt with markdown docs and a 356-operation OpenAPI. - [Calendly](https://agentdrives.com/tools/calendly.md?via=llms): L1 · Needs a person, Scheduling. Calendly documents its API well for agents (llms.txt, Markdown pages, OpenAPI 3.1, hosted MCP with OAuth and DCR), but creating an account stops at a visible reCAPTCHA v2 after the sign-up form. - [Canva](https://agentdrives.com/tools/canva.md?via=llms): L1 · Needs a person, Design. An agent cannot get into Canva: every www.canva.com page, including sign-up and the OAuth consent page that REST, MCP and the CLI all need, shows a Cloudflare interstitial that never cleared for our agent's browser. - [Chatwoot](https://agentdrives.com/tools/chatwoot.md?via=llms): L1 · Needs a person, Customer support. Chatwoot's developer surface is strong for agents: llms.txt, Markdown pages, OpenAPI 3.1.0, an official CLI with an agent skill, and a modern docs MCP. - [ClickUp](https://agentdrives.com/tools/clickup.md?via=llms): L1 · Needs a person, Project management. ClickUp has strong agent-facing docs (llms.txt, markdown pages, OpenAPI v2 and v3) and an official remote MCP server with OAuth discovery, but an agent cannot get in on its own: the only email sign-up shows a visible reCAPTCHA v2 checkbox. - [Close](https://agentdrives.com/tools/close.md?via=llms): L1 · Needs a person, CRM. Close's sign-up refused our agent's browser through an invisible reCAPTCHA check, with no challenge a human could solve, so we never got an account, key or token. - [Crisp](https://agentdrives.com/tools/crisp.md?via=llms): L1 · Needs a person, Customer support. Crisp has a broad REST API, official SDKs and an official MCP server, but an agent cannot get in on its own: the workspace sign-up shows a visible hCaptcha puzzle and every working token is issued inside the workspace dashboard. - [DocuSign](https://agentdrives.com/tools/docusign.md?via=llms): L1 · Needs a person, E-signature. Docusign's docs, Swagger specs and MCP auth metadata are public, but both ways in (the free developer account and the free trial) stop at an Arkose Labs puzzle, so our agent never got an account. - [Dropbox](https://agentdrives.com/tools/dropbox.md?via=llms): L1 · Needs a person, File storage. Dropbox's only sign-up is a web form that shows an Arkose Labs puzzle, so an agent cannot get an account on its own and no capability could be tried. - [Ecwid](https://agentdrives.com/tools/ecwid.md?via=llms): L1 · Needs a person, E-commerce. An agent can read Ecwid's developer docs well (llms.txt, Markdown pages, a docs-search MCP), but it cannot open a store: the sign-up form shows a visible Cloudflare Turnstile checkbox, and there is no API sign-up. - [Fathom Analytics](https://agentdrives.com/tools/usefathom.md?via=llms): L1 · Needs a person, Web analytics. Fathom Analytics has a clean, agent-friendly API surface (llms.txt, Markdown API reference, OpenAPI 3.1.0), but an agent cannot get in: the only way to start is a 7-day trial whose sign-up form requires a credit card and ZIP code, and API tokens are made in the web app. - [Figma](https://agentdrives.com/tools/figma.md?via=llms): L1 · Needs a person, Design. An agent cannot get into Figma on its own: robots.txt disallows /signup, /login and the MCP consent page for AI agents that act for a user, and the MCP server's client registration answers 403 to clients outside Figma's catalog. - [folk](https://agentdrives.com/tools/folk.md?via=llms): L1 · Needs a person, CRM. folk documents its API and official MCP server well (llms.txt, Markdown pages, OpenAPI 3.1.0, PRM and CIMD/DCR on the MCP OAuth). - [Formspree](https://agentdrives.com/tools/formspree.md?via=llms): L1 · Needs a person, Forms. An agent cannot get into Formspree on its own: sign-up is a web form, and Register raised a reCAPTCHA v2 picture puzzle. - [Framer](https://agentdrives.com/tools/framer.md?via=llms): L1 · Needs a person, Website builders. Framer's docs are very agent-readable (llms.txt, markdown on every page) and it offers a CLI, a Server API package and an OAuth MCP endpoint. - [FreshBooks](https://agentdrives.com/tools/freshbooks.md?via=llms): L1 · Needs a person, Invoicing. An agent cannot get into FreshBooks alone: the sign-up page on auth.freshbooks.com showed our agent's browser a visible reCAPTCHA image puzzle, so no capability was driven. - [Freshdesk](https://agentdrives.com/tools/freshdesk.md?via=llms): L1 · Needs a person, Customer support. Freshdesk has a documented REST API and an official MCP server (API-key auth, Growth plan and up), but an agent cannot get an account: the sign-up page on www.freshworks.com is behind a Cloudflare managed challenge that our signed agent did not pass. - [Front](https://agentdrives.com/tools/front.md?via=llms): L1 · Needs a person, Customer support. Front's docs are well prepared for agents (llms.txt, Markdown pages, OpenAPI, an official MCP server), but an agent cannot get in: the only sign-up route is a web form that requires a phone number, and API tokens and MCP OAuth clients are only made inside an account. - [Gorgias](https://agentdrives.com/tools/gorgias.md?via=llms): L1 · Needs a person, Customer support. An agent cannot open a Gorgias account: the sign-up form's invisible Cloudflare Turnstile never cleared for our agent's browser, and no API sign-up exists. - [Heap](https://agentdrives.com/tools/heap.md?via=llms): L1 · Needs a person, Product analytics. Heap's developer docs suit agents: an llms.txt, Markdown pages and an OpenAPI 3.1.0 definition per reference page. - [Help Scout](https://agentdrives.com/tools/helpscout.md?via=llms): L1 · Needs a person, Customer support. Help Scout's sign-up form accepts an agent's address without a bot check, but then requires an SMS code to a phone number before the account exists, so an agent alone gets no account, no key and no API or MCP call. - [HubSpot](https://agentdrives.com/tools/hubspot.md?via=llms): L1 · Needs a person, CRM. HubSpot's docs are agent-friendly (llms.txt, markdown pages, an official MCP server and an Agent CLI), but an agent cannot get in on its own: the email sign-up shows a visible reCAPTCHA checkbox before the code is sent. - [Invoice Ninja](https://agentdrives.com/tools/invoiceninja.md?via=llms): L1 · Needs a person, Invoicing. Invoice Ninja has a broad, well-documented REST API (OpenAPI, 365 operations) and the production API answers agents from a datacenter without bot checks, but an agent cannot get an account: the only sign-up is a web form with a visible Cloudflare Turnstile checkbox. - [Jira](https://agentdrives.com/tools/atlassian.md?via=llms): L1 · Needs a person, Project management. An agent cannot get into Jira on its own: both email sign-ups are stopped by reCAPTCHA (an invisible score refusal, then a visible checkbox). - [Jotform](https://agentdrives.com/tools/jotform.md?via=llms): L1 · Needs a person, Forms. Jotform has a broad documented REST API, SDKs and an official OAuth MCP server, but our agent never got credentials. - [Kit](https://agentdrives.com/tools/kit.md?via=llms): L1 · Needs a person, Email marketing. Kit's developer docs are agent-friendly (llms.txt, Markdown pages, OpenAPI 3.0.3, a docs MCP on the 28 Jul 2026 spec), but app.kit.com/users/signup and all of kit.com answer our signed agent with a Cloudflare 'Verify you are human' challenge, so no account, key or core task was reached. - [Klaviyo](https://agentdrives.com/tools/klaviyo.md?via=llms): L1 · Needs a person, Email marketing. Klaviyo has a strong agent surface on paper: a 345-operation REST API with an OpenAPI spec, Markdown docs, an llms.txt, and an official MCP server (remote with OAuth/CIMD, local with an API key). - [Lemon Squeezy](https://agentdrives.com/tools/lemonsqueezy.md?via=llms): L1 · Needs a person, Payments. An agent cannot sign up for Lemon Squeezy: the registration form needs a reCAPTCHA v2 checkbox, and there is no API sign-up. - [Linear](https://agentdrives.com/tools/linear.md?via=llms): L1 · Needs a person, Project management. An agent cannot get into Linear on its own: the email sign-up shows a visible Cloudflare Turnstile and no other way in fits an agent (Google needs an account we lack, no GitHub, no API sign-up). - [Mailchimp](https://agentdrives.com/tools/mailchimp.md?via=llms): L1 · Needs a person, Email marketing. Mailchimp's developer docs are agent-friendly: llms.txt, Markdown pages, agent skills and a docs-search MCP (legacy, its tool failing). - [MailerLite](https://agentdrives.com/tools/mailerlite.md?via=llms): L1 · Needs a person, Email marketing. MailerLite has a strong agent surface on paper: Markdown docs with llms.txt, an official CLI, six SDKs and a hosted OAuth MCP server with 77 tools. - [Mailgun](https://agentdrives.com/tools/mailgun.md?via=llms): L1 · Needs a person, Email infrastructure. An agent cannot get into Mailgun: the free sign-up form was rejected by an invisible reCAPTCHA ("Could not validate reCaptcha response") and there is no API sign-up. - [Mailtrap](https://agentdrives.com/tools/mailtrap.md?via=llms): L1 · Needs a person, Email infrastructure. Mailtrap has strong agent surfaces on paper: llms.txt and Markdown docs, OpenAPI 3.1.0 (136 operations), an official CLI and a stdio MCP server with 125 tools. - [Make](https://agentdrives.com/tools/make.md?via=llms): L1 · Needs a person, Automation. Make documents a full REST API, CLI, SDK and hosted MCP server, and its docs are agent-friendly (llms.txt, markdown). - [Mollie](https://agentdrives.com/tools/mollie.md?via=llms): L1 · Needs a person, Payments. An agent cannot open a Mollie account: the sign-up form's reCAPTCHA rejected our agent's browser and showed a picture puzzle. - [Moneybird](https://agentdrives.com/tools/moneybird.md?via=llms): L1 · Needs a person, Invoicing. Moneybird has unusually good agent tooling on paper: an official MCP server with OAuth (PRM, DCR and CIMD), an official CLI, a 297-operation OpenAPI 3.1 spec and llms.txt. - [MongoDB Atlas](https://agentdrives.com/tools/mongodb.md?via=llms): L1 · Needs a person, Databases. An agent cannot get into MongoDB Atlas on its own: the sign-up form's invisible reCAPTCHA rejected our signed agent (INVALID_CAPTCHA, shown as a 'server error'), and robots.txt disallows the Atlas CLI's sign-up page. - [n8n](https://agentdrives.com/tools/n8n.md?via=llms): L1 · Needs a person, Automation. n8n's docs are agent-friendly (llms.txt, Markdown pages, a docs MCP server) and the product offers an official MCP server, a public API and a CLI. - [Notion](https://agentdrives.com/tools/notion.md?via=llms): L1 · Needs a person, Docs and workspace. Notion has an excellent agent surface on paper: REST API with OpenAPI, the ntn CLI, an official hosted MCP server with dynamic client registration, webhooks and markdown docs with llms.txt. - [PandaDoc](https://agentdrives.com/tools/pandadoc.md?via=llms): L1 · Needs a person, E-signature. PandaDoc's developer docs are among the most agent-ready we have seen (llms.txt, markdown pages, OpenAPI, a keyless docs MCP and an official OAuth MCP server), but an agent cannot get in: the free trial sign-up passes both form steps and then its invisible reCAPTCHA answers "Captcha is not valid" to our agent's browser. - [Perplexity API](https://agentdrives.com/tools/perplexity.md?via=llms): L1 · Needs a person, Search APIs. Perplexity's docs are agent-ready (llms.txt, markdown pages, six OpenAPI specs, an official hosted MCP server with RFC 9728 metadata and CIMD), but an agent cannot get in: console.perplexity.ai and www.perplexity.ai answer our signed agent with a Cloudflare 'Verify you are human' challenge, also in a real browser. - [Pinata](https://agentdrives.com/tools/pinata.md?via=llms): L1 · Needs a person, File storage. Pinata's docs are very agent-friendly (llms.txt, markdown, a modern docs MCP), and it runs a first-party MPP endpoint that sells account-free IPFS uploads and downloads. - [Pirsch](https://agentdrives.com/tools/pirsch.md?via=llms): L1 · Needs a person, Web analytics. Pirsch has a large, well documented REST API (about 150 operations, including account delete) and a docs-linked third-party MCP server that speaks MCP (the 28 Jul 2026 spec), but an agent cannot get in: sign-up is a web form with an ALTCHA 'I'm not a robot' checkbox and there is no sign-up API. - [Plain](https://agentdrives.com/tools/plain.md?via=llms): L1 · Needs a person, Customer support. Plain has a public GraphQL schema, SDK, CLI, an official OAuth MCP server and markdown docs, but an agent cannot create an account: the WorkOS sign-up page shows a visible Cloudflare Turnstile and the email sign-in path answered 'Access blocked'. - [Polar](https://agentdrives.com/tools/polar.md?via=llms): L1 · Needs a person, Payments. Polar's docs are agent-ready (llms.txt, Markdown pages, OpenAPI 3.1.0, CLI, OAuth MCP server), but our agent could not get in: email sign-in sits behind a Cloudflare Turnstile widget that failed our agent's browser. - [PostHog](https://agentdrives.com/tools/posthog.md?via=llms): L1 · Needs a person, Product analytics. PostHog is very well documented for agents: llms.txt, Markdown for every page, a public OpenAPI spec, an MCP server with OAuth discovery and an auth.md manifest. - [Postmark](https://agentdrives.com/tools/postmarkapp.md?via=llms): L1 · Needs a person, Email infrastructure. Postmark's agent-facing docs are unusually complete (llms.txt, AI catalog, auth.md, api-catalog, Agent Skills, MCP server card) and the test token lets an agent dry-run sends with no account. - [Publer](https://agentdrives.com/tools/publer.md?via=llms): L1 · Needs a person, Social media scheduling. An agent cannot get into Publer: the only sign-up is a web form whose reCAPTCHA showed our agent's browser a picture puzzle. - [Render](https://agentdrives.com/tools/render.md?via=llms): L1 · Needs a person, Hosting and deploys. Render's agent docs are excellent: llms.txt, Markdown for every page, a docs-search API and a hosted MCP server. - [Resend](https://agentdrives.com/tools/resend.md?via=llms): L1 · Needs a person, Email infrastructure. Resend publishes agent docs (llms.txt, markdown pages, OpenAPI, auth.md, skills, CLI, OAuth MCP server), but our agent could not get an account: the sign-up form returned a generic server error twice, with no visible challenge to hand to a human. - [Salesforce](https://agentdrives.com/tools/salesforce.md?via=llms): L1 · Needs a person, CRM. An agent cannot get into Salesforce: the free Developer Edition sign-up on developer.salesforce.com and the Trailblazer sign-in behind the Starter Suite trial are refused by Akamai (403, no challenge), and the Enterprise trial form requires a phone number. - [SavvyCal](https://agentdrives.com/tools/savvycal.md?via=llms): L1 · Needs a person, Scheduling. SavvyCal documents a REST API, but an agent cannot get in: the sign-up form shows a visible hCaptcha puzzle. - [ScrapingBee](https://agentdrives.com/tools/scrapingbee.md?via=llms): L1 · Needs a person, Web scraping. ScrapingBee's only sign-up is a web form with a visible reCAPTCHA v2 checkbox, so an agent cannot get an account or API key alone. - [SendGrid](https://agentdrives.com/tools/sendgrid.md?via=llms): L1 · Needs a person, Email infrastructure. An agent can read all of SendGrid's docs headless (llms.txt, Markdown pages, OpenAPI 3.1.0, a docs-only MCP server) but cannot get an account: the Twilio sign-up form shows a Cloudflare Turnstile check to take it. - [SerpApi](https://agentdrives.com/tools/serpapi.md?via=llms): L1 · Needs a person, Search APIs. SerpApi has agent-friendly docs (llms.txt, Markdown pages) and a current, stateless MCP server, but an agent cannot get an account: the sign-up form shows a reCAPTCHA picture puzzle and there is no API sign-up. - [Slack](https://agentdrives.com/tools/slack.md?via=llms): L1 · Needs a person, Team chat. An agent cannot get into Slack on its own: the email sign-up and the Developer Program join form both show a visible reCAPTCHA 'I'm not a robot' checkbox to tick it, so no account, token or core task was reached. - [Squarespace](https://agentdrives.com/tools/squarespace.md?via=llms): L1 · Needs a person, Website builders. An agent cannot create a Squarespace account: the web-only sign-up, guarded by invisible reCAPTCHA Enterprise, returned a 401 'username or password is incorrect' and showed nothing. - [Taskade](https://agentdrives.com/tools/taskade.md?via=llms): L1 · Needs a person, Task management. An agent cannot get into Taskade on its own: the only sign-up is a web form with a visible Cloudflare Turnstile widget to do it. - [tawk.to](https://agentdrives.com/tools/tawk.md?via=llms): L1 · Needs a person, Customer support. An agent can read tawk.to's docs (llms.txt, developer portal, help center), but the only sign-up route is a web form behind Cloudflare Turnstile. - [Teamwork.com](https://agentdrives.com/tools/teamwork.md?via=llms): L1 · Needs a person, Project management. Teamwork.com has an official MCP server that speaks MCP (the 28 Jul 2026 spec), OAuth metadata and a 429-operation Swagger spec, but an agent cannot get in: the sign-up form requires a phone number and there is no headless sign-up or credential path. - [Todoist](https://agentdrives.com/tools/todoist.md?via=llms): L1 · Needs a person, Task management. Todoist has a strong agent surface on paper: an OpenAPI 3.1.0 REST API, an official OAuth MCP server with PRM and CIMD/DCR, a CLI and SDK. - [Trello](https://agentdrives.com/tools/trello.md?via=llms): L1 · Needs a person, Project management. Trello has a full REST API (OpenAPI 3.0.0, 261 operations) and an official remote MCP server with current OAuth discovery (PRM, CIMD), but both need a signed-in Atlassian account. - [Turso](https://agentdrives.com/tools/turso.md?via=llms): L1 · Needs a person, Databases. Turso's docs are very agent-friendly (llms.txt, Markdown pages, OpenAPI, a hosted OAuth MCP server with CIMD), and its embedded engine runs locally through the official SDK with no account: SQL, vectors, FTS and CDC all worked. - [Typeform](https://agentdrives.com/tools/typeform.md?via=llms): L1 · Needs a person, Forms. Typeform has a documented REST API and an official MCP server with proper OAuth discovery (PRM, CIMD, DCR), but an agent cannot get an account: our agent's browser sign-up was silently denied by Typeform's registration inline hook (E0000135) with a misleading 'you might already have an account' message. - [Typefully](https://agentdrives.com/tools/typefully.md?via=llms): L1 · Needs a person, Social media scheduling. Typefully has an agent-ready surface: an OpenAPI 3.1.0 REST API, an official MCP server with OAuth (CIMD) or API-key auth, llms.txt and Markdown docs, and API and MCP on the free plan. - [Wix](https://agentdrives.com/tools/wix.md?via=llms): L1 · Needs a person, Website builders. Wix documents a broad REST API, SDK, CLI and an official MCP server, with markdown docs for agents. - [Wrike](https://agentdrives.com/tools/wrike.md?via=llms): L1 · Needs a person, Project management. Wrike's docs are agent-ready: llms.txt, Markdown pages, an api-catalog with OpenAPI 3.0.1, and an official MCP server with Protected Resource Metadata. - [Xero](https://agentdrives.com/tools/xero.md?via=llms): L1 · Needs a person, Invoicing. Xero stops an agent at sign-up: both the free trial and the free developer login require a phone number, so no account, credentials or capability could be reached. - [Zendesk](https://agentdrives.com/tools/zendesk.md?via=llms): L1 · Needs a person, Customer support. Zendesk's free 14-day trial needs no card, but its sign-up form requires a phone number, so an agent cannot create an account; the only API route to create accounts is reseller-only. - [Carrd](https://agentdrives.com/tools/carrd.md?via=llms): L0 · Read-only, Website builders. Carrd has no API, CLI or MCP: everything runs in the web builder. - [Later](https://agentdrives.com/tools/later.md?via=llms): L0 · Read-only, Social media scheduling. An agent cannot get into Later Social: app.later.com, which serves sign-up, log-in and checkout, shows our signed agent a Cloudflare Turnstile check. - [zcal](https://agentdrives.com/tools/zcal.md?via=llms): L0 · Read-only, Scheduling. zcal has no public API, CLI or MCP server; everything happens in its web app. ## Fixes One guide per wall our agent meets, for owners, with the fix in steps and a check to run. The list as Markdown: https://agentdrives.com/fixes.md - [Let signed AI agents past your bot check](https://agentdrives.com/fixes/let-signed-agents-past-your-bot-check.md): The setting for Cloudflare, Vercel, Akamai, DataDome and HUMAN, and what to do about a check on the form. - [Let AI agents sign up and get an API key](https://agentdrives.com/fixes/let-agents-sign-up-and-get-a-key.md): An API call that signs up and returns a limited key, and a claim step for the person. - [MCP sign-in an AI agent can finish: OAuth and CIMD](https://agentdrives.com/fixes/mcp-sign-in-agents-can-finish.md): OAuth metadata, CIMD or DCR, and a way to approve without a browser on the agent's side. - [Let AI agents start without a credit card](https://agentdrives.com/fixes/card-free-starts.md): A free plan or a trial without a card, an API key that works on it, and limits instead of a card. - [Let AI agents pay you per call: x402 and MPP](https://agentdrives.com/fixes/let-agents-pay-per-call.md): x402 or MPP on a paid endpoint, your prices where agents read them, and plans an agent can buy by API or MCP. - [Docs AI agents can read: llms.txt, Markdown, OpenAPI](https://agentdrives.com/fixes/docs-agents-can-read.md): An llms.txt, Markdown twins of your pages, an OpenAPI spec and an MCP server card. ## Research - [Sign-up is the wall](https://agentdrives.com/findings/the-front-door?via=llms): 88 of 140 products stop an AI agent at the door: it can't get in on its own. It got into the other 52. Drives up to 16 Oct 2026. - [MCP servers everywhere, agents nowhere](https://agentdrives.com/findings/mcp-without-access?via=llms): 76 of the 124 products with an MCP server keep an agent out: it can't get in on its own to use the server. - [Agents can pay, but only next to the product](https://agentdrives.com/findings/agents-pay-at-the-side-door?via=llms): 1 of the 5 products that take agent-native payment don't take it for the product you sign up for. - [Agents can't leave](https://agentdrives.com/findings/no-way-out?via=llms): 17 of the 44 products where our agent got as far as deleting its account gave it no way to do it on its own. - [Bot checks, provider by provider](https://agentdrives.com/research/bot-checks?via=llms) - [Which project-management tools can agents run? Q4 2026](https://agentdrives.com/research/indices/project-management.md?via=llms): 12 project-management tools ranked by level, driven 2–16 Oct 2026. Data: https://agentdrives.com/research/indices/project-management.csv?via=llms, https://agentdrives.com/research/indices/project-management.json?via=llms - [Which customer support tools can agents run? Q4 2026](https://agentdrives.com/research/indices/customer-support.md?via=llms): 10 customer support tools ranked by level, driven 6–15 Oct 2026. Data: https://agentdrives.com/research/indices/customer-support.csv?via=llms, https://agentdrives.com/research/indices/customer-support.json?via=llms - [State of agent access, October 2026](https://agentdrives.com/research/state-of-agent-access-2026-10?via=llms): what an AI agent met at the 5,000 most-linked software sites, 12–13 Oct 2026. Data: https://agentdrives.com/research/state-of-agent-access-2026-10.json?via=llms ## Pages - [How we test](https://agentdrives.com/method?via=llms) - [Research](https://agentdrives.com/research?via=llms) - [Example report](https://agentdrives.com/example-report?via=llms) - [Pricing](https://agentdrives.com/pricing?via=llms) and [the offer as JSON](https://agentdrives.com/pricing.json?via=llms) - [Facts: every number with its evidence](https://agentdrives.com/data/facts?via=llms) - [Contact](https://agentdrives.com/contact?via=llms)