Can an AI agent use Mistral AI?
Mistral AI sells AI models and tools for building agents.
Rank among the 246 products we drove
- Steps on its own
- tied #3
- 4/ 5
- Find to Leave, no person
- Core work done
- tied #1
- 100%
- 10 of 10 core tasks
- Time to get in
- tied #61
- 5minutes
- Getting in on its own
Yes. Our agent signed up for Mistral AI on its own in a browser through a device login, got an API key and used its API, MCP server and CLI.
The record
Level confirmed
Not yet
Level from one drive; a second must agree.
- Got in by
- Sign-up in a browser5 minutes
- Got as far as
- Pay: reaches the checkoutA card form
Connects by
MCP, API, CLI, SDK
Agent payments
None found
On its own · 505 doc pages read
Docs it can act on: llms.txt, an OpenAPI and an MCP server.
On its own · 5 minutes
Sign-up in a browser.
On its own · 10 of 10 core tasks
Did: get chat answers, embeddings and moderation through the API; create, change, use and delete custom agents and conversations.
Partly
Reaches the checkout. A card form.
On its own
Nothing to cancel. It deletes its account.
- On its own
- Partly
To reach L5: Pay without a person.
See how it ranks in the model APIs category.
Own Mistral AI? 10 things to fix. Get your findings, free
Checked 18 Nov 2026: MCP server asks for a sign-in, OpenAPI changed since the check before, llms.txt unchanged since the check before.
Find
How an agent connects
MCP, API, OpenAPI, CLI, SDK and llms.txt found.
MCP server
60 of 62 testedOfficial server. Remote server https://api.mistral.ai/mcp, MCP (the 28 Jul 2026 spec), 62 tools when driven, Bearer API key. Sign‑in needed to see its tools.
62 tools mapped in our drive
API
196 of 321 testedREST, OpenAPI 3.1.0 at docs.mistral.ai/openapi.yaml, 321 operations.
321 operations mapped in our drive
OpenAPI
YesA machine-readable description of the API.
CLI
All testedVibe CLI 3.0.4 (browser-free -p mode, browser sign-in); mistralai-workflows-cli 1.3.0.
13 commands mapped in our drive
SDK
All tested10 methods mapped in our drive
llms.txt
YesAgent payments
None foundReaches the checkout: a card form.
Docs
Some gapsCovers getting started, auth, error codes and code examples.
Describes each operation only in part: completeness 57 of 100.
11 of 11 documented ways to do the core work worked.
505 of 627 pages read
Changed since it was driven
Checked daily
- OpenAPI: 7 endpoints added and 9 removed
Get in
Getting in
Our agent tried 2 of 4 ways in; 2 worked without a person.
The way that worked
Sign-up in a browser5 minutes
- An inboxNeeded
- A browserNot needed
- A personNot needed
- Agent on a serverAPI, MCP, CLINoStops at sign-up: it works only in a browser.
- Assistant in its own browserLike ChatGPT agentPartlySigns up, confirms its email and does all of the core work. Payment: reaches the checkout, a card form.
- Signed agentKnown to CloudflareNoTurned away before the sign-up page.
Every way in
| Way in | What happened | Why |
|---|---|---|
| Sign-up in a browser | Worked | |
| Sign-up in a browser | Worked | Part of the work done |
| Company single sign-on | Not tried | Needs a person's account |
| A key from a person | Not tried | Needs a person's account |
Use
What an agent can do with Mistral AI
All 10 core tasks done, and 15 of 21 others.
Core tasks
- done there
| Task | Result | Through |
|---|---|---|
| Get a chat completion from a Mistral model | Done | API: done thereMCP: done there |
| Create a custom AI agent | Done | API: done there |
| Chat with an agent in a conversation | Done | API: done there |
| Extract text and structure from a document with OCR | Done | API: done there |
| Transcribe an audio file | Done | API: done there |
| Generate speech from text | Done | API: done there |
| Create a document library and add documents for RAG | Done | API: done thereMCP: done there |
| Build and run a durable workflow | Done | API: done thereMCP: done there |
| Give Vibe an everyday task to do on its own | Done | Web app: done there |
| Run a coding session with the Vibe coding agent | Done | CLI: done there |
Surface audit: 418 documented operations: 284 run, 96 blocked by sandbox or plan limits and 38 left out because they spend money or reach real people. Of the 284 run, 236 passed and 48 didn't; 23 behaved differently from the docs. By surface, 196 on the API, 60 on the MCP server, 13 in the CLI, 10 in the SDK and 5 elsewhere, such as sign-in and webhooks.
Holds up
68of 100
- Reliability88
- Handles mistakes48
The mean of the two.
Docs match
81of 100
- Docs accuracy92
- Docs completeness57
Docs accuracy counts 70%, completeness 30%.
Surface tested
62of 100
236 of 380 passed: the 418 documented operations less the 38 that spend money or reach real people
What it can't do, and why
- Create a working API key from the Studio key page on Free The 'New key' button is disabled and the Admin-panel key gets 0 requests per minute.
- Run batch jobs The API answers 402 on the Free plan.
- Search traces or run evaluations Observability is Enterprise only; the API answers a bare 404.
- Clone a custom voice A paid plan is required.
- Make native slides or spreadsheets in Vibe These are Pro features; Free gives an HTML canvas.
- Delete the account through the API No delete API was found; deleting works in the browser only.
15 of 21 other tasks done
- Create text or code embeddingsDone
- Save a reusable prompt with versionsDone
- Create a SkillDone
- Add an app connector (e.g. Google, Notion) for agents to useDone
- Moderate text for harmful contentDone
- Upload a file for fine-tuning, batch or OCRDone
- Schedule a workflow to runDone
- Generate an image with AIDone
- Schedule a recurring task in VibeDone
- Organise chats in a project folderDone
- Run deep researchDone
- Store a secret for workflows and connectorsDone
- Create and search a managed search indexDone
- Read usage and billingDone
- Export account dataDone
- Create slides in VibePartly
- Run a batch inference jobPaid plan
- Search traces of an AI applicationPaid plan
- Run an evaluation on a datasetPaid plan
- Clone a custom voiceA paid plan is required.Paid plan
- Create a spreadsheet in VibePaid plan
- Configure SAML SSOPaid plan
- Invite a member to the organizationBlocked
- Create a workspaceBlocked
- Create a user groupBlocked
- Set a spending limitBlocked
- Read the audit logBlocked
The plans they need
- credentialsPay-as-you-go or Pro
- The first callPay-as-you-go or Pro
- Export account dataTeam
The drive used the free plan.
Tips for agent builders
- On the Free plan, run
vibe --setup, choose Launch browser and authorize once: the Vibe-type key it stores (OS keyring service ai.mistral.vibe) works for the REST API and the MCP server. - Sign up at console.mistral.ai, not cloud.mistral.ai: the latter blocks agent browsers with Cloudflare.
- If a call answers 429 'Rate limit exceeded' with x-ratelimit-limit-req-minute: 0, the key type is the problem: switch to the Vibe-sign-in key instead of backing off.
- Workflows workers need a Studio-type key. The Vibe key gets 401 at the workflow scheduler, and
make start-workerhides that error, so run entrypoints.worker directly to see it. - Send Idempotency-Key with care: POST /v1/agents creates a duplicate on a retry anyway, so list before you retry creates.
- Use the MCP server at https://api.mistral.ai/mcp with Authorization: Bearer <key>; it speaks MCP 2026-07-28 (server/discover) and needs no OAuth.
Pay
Paying
Ways to pay checked 2 Nov 2026.
- ResultPartly. Reaches the checkout. A card form.
- Ways to pay
- MPP (a third party's · listed or named only)
- TrialIt has a free plan.
Leave
Leaving
- ResultOn its own. Nothing to cancel. It deletes its account.
- Its accountYes, in a browser, without a human. The account and its single-member organization went: Vibe > Settings > Profile > Delete account, typing the email to confirm, after a forced re-sign-in (password, no CAPTCHA). Afterwards the key returned 401, password login was refused and the old session returned 401. This works in the browser only; no API delete exists.
Daily checks
Checked every day
What our agent finds on the product's own site each day, without an account.
Sign-up page,
No sign-up page on the newest check; 16 of 16 days
llms.txt,
Found on the newest check; 16 of 16 days
MCP server,
Asks for a sign-in on the newest check; 4 of 16 days
OpenAPI,
Found on the newest check; 4 of 16 days
Compare
How Mistral AI compares
In Model APIs
Mistral AI is the highest of the 2 products we drove in Model APIs. As of 15 Nov 2026.
| Product | Level | Find to leave |
|---|---|---|
| Mistral AI (this page) | L4 | Find: On its own, Get in: On its own, Use: On its own, Pay: Partly, Leave: On its own. |
| Agentgates | L2 | Find: Partly, Get in: On its own, Use: Not measured, Pay: On its own, Leave: Partly. |
For the team behind Mistral AI
Agents get stuck at Pay
10 things to fix for agents: 5 small, 4 medium and 1 large
One fix can cover several findings.
Agent blockers: 8 foundAgent surfaces: 24 issues found
Get your findings, freeWhat a drive covers
We mail a private link to an address at mistral.ai.
It opens the two findings that matter most in full, and every other by title. There is nothing to sign in to.
- FindClear
- Get inClear
- UseClear
- PayBlocks agents
- LeaveClear
Or claim Mistral AI to see where each issue is and ask for a free drive after a fix.
Want Mistral AI driven every month? See Watch
On our standard
Best or ahead on 9 of 37 criteria, behind on 3. Each against every product we measured on it.
- Spend an agent can see and capAhead of 153 of 159 others, 6 the same.
- Typed, tight input schemasAhead of 77 of 91 others, 14 the same.
- All validation problems at onceAhead of 47 of 57 others, 10 the same.
- Export before leavingAhead of 148 of 189 others, 41 the same.
- Pages fit one fetchLevel with 118 of 187 others, 69 higher.
- Answers sized for an agentBehind 86 of 116 others, 13 the same, 17 lower.
- Rate limits an agent can obeyBehind 10 of 14 others, 4 the same.
- Agents may read docs, API and sign-upBehind 130 of 160 others, 30 the same.
All 37 criteria
- llms.txt that worksAhead of 185 of 239 others, 54 the same.
- Structured resultsAhead of 55 of 78 others, 23 the same.
- Every way in works for an agentAhead of 134 of 193 others, 59 the same.
- A way in with no personAhead of 137 of 243 others, 106 the same.
- Sign-in for agents acting for a personLevel with 83 of 158 others, 75 lower.
- Versions an agent can pinLevel with 73 of 127 others, 54 lower.
- Delete by API, checkedIn the middle 46 of 97 others lower, 43 the same, 8 higher.
- One way to page through listsIn the middle 49 of 114 others lower, 52 the same, 13 higher.
- Prices an agent can readLevel with 117 of 153 others, 36 lower.
- Errors a program can readLevel with 82 of 145 others, 47 lower, 16 higher.
- No person in the buy pathLevel with 114 of 138 others, 24 lower.
- Limits known before they hitAhead of 48 of 90 others, 9 the same, 33 higher.
- Fast to the first callLevel with 118 of 135 others, 17 lower.
- Descriptions that say when to use the toolLevel with 56 of 91 others, 22 lower, 13 higher.
- The best way in is built for agentsIn the middle 79 of 196 others lower, 55 the same, 62 higher.
- Tool names that sort themselvesLevel with 85 of 91 others, 6 lower.
- Terms allow an agent's accountLevel with 179 of 189 others, 10 lower.
- Real status codesLevel with 161 of 168 others, 7 lower.
- Errors say what to do nextLevel with 107 of 145 others, 22 lower, 16 higher.
- A path written for agentsIn the middle 48 of 187 others lower, 91 the same, 48 higher.
- WebMCP tools on the web appLevel with 133 of 136 others, 3 higher.
- An agent payment rail on the coreLevel with 205 of 216 others, 11 higher.
- A plan an agent can buy by APILevel with 157 of 170 others, 13 higher.
- A request id on every answerIn the middle 42 of 143 others lower, 47 the same, 54 higher.
- Results sized for a context windowIn the middle 16 of 70 others lower, 27 the same, 27 higher.
- Truthful annotationsIn the middle 24 of 91 others lower, 25 the same, 42 higher.
- Agents pick the right toolLevel with 72 of 91 others, 19 higher.
- Docs an agent can navigateLevel with 108 of 187 others, 19 lower, 60 higher.
- A tool list that fitsIn the middle 21 of 91 others lower, 26 the same, 44 higher.
Where Mistral AI fits
- Tasks
- Standards
- Topics
- MCP serversFind step
- Docs and llms.txtFind step
- MCP sign-inGet in step
- Sign-upGet in step
- MCP and APIUse step
- Card checkoutPay step
- Delete the accountLeave step
- Rivals
History
Level by drive
One drive so far, on 2 Nov 2026.
One drive so far.
Read more on this step
Pay is where our agent stopped at Mistral AI.
- Finding · 18 Nov 2026Can an AI agent pay with x402 or MPP?4 of the 12 products that offer x402 or MPP took a payment from our agent on their drive.Read
- Guide · 9 Nov 2026Agent experience (AX): what it is, measured at 246 productsAs of 9 Nov 2026: Agent experience (AX) is whether an AI agent can find, get into, use and pay for a product on its own. We drove 246 products to measure it.Read
- Guide · 31 Oct 2026Bot checks on SaaS checkouts: where AI agents stopAs of 31 Oct 2026: 42 of 59 SaaS checkouts our agent reached on 31 Oct 2026 carried a bot check; under 5 made it solve one. 80 of 139 stopped earlier. What to fix so an agent can buy.Read
Every earlier drive is in the history. The level follows these rules. Data under CC BY 4.0.
Found an error? Tell us on this page or read how corrections work.
Questions people ask about Mistral AI
- Can an AI agent create a Mistral AI account on its own?
- Yes, in a browser, without a person. The sign-up form at console.mistral.ai (through v2.auth.mistral.ai) takes email, password and name, and showed no bot check. The agent's account was made at that auth service and carried the whole drive. The agent called the service's own flow directly, which is not a documented route. Name asked; our agent answered. The side host cloud.mistral.ai showed a Cloudflare block page for our agent's browser, but it has no sign-up form and did not stop sign-up.
- Does Mistral AI have an MCP server, and can an agent connect to it?
- Yes. Mistral AI has an official MCP server. The server is hosted, lists 62 tools and speaks the 28 Jul 2026 MCP spec. A new agent can get in on its own: in the drive of 2 Nov 2026, our agent signed up, verified and got its key without a person. Through the MCP server, our agent tried 6 tasks and did 6.
- How does an AI agent get a Mistral AI API key?
- In the drive of 2 Nov 2026, our agent got its Mistral AI API key in a browser, without a person. Studio's own /api-keys page disables 'New key' on Free ('Upgrade to activate your API keys'). A key made in the Admin panel got 0 requests per minute on inference. The working key came from
vibe --setup> Launch browser: one Authorize click in the agent's own signed-in browser provisioned a Vibe-type key that works for REST, MCP and the CLI. The Admin-panel key was revoked in the browser (401 afterwards). No documented browser-free route makes the first key. - Can an AI agent delete its Mistral AI account?
- Yes, in a browser, without a human. The account and its single-member organization went: Vibe > Settings > Profile > Delete account, typing the email to confirm, after a forced re-sign-in (password, no CAPTCHA). Afterwards the key returned 401, password login was refused and the old session returned 401. This works in the browser only; no API delete exists.
- Can an AI agent pay for Mistral AI?
- Partly. Our agent reached a card form. It has a free plan.
Cite this page
The page as a reference, in plain text, APA and BibTeX.
Cite this
AgentDrives, “Mistral AI: AI agent test drive”, 2 November 2026, https://agentdrives.com/tools/mistral. CC BY 4.0.
AgentDrives. (2026, November 2). Mistral AI: AI agent test drive. https://agentdrives.com/tools/mistral
@misc{agentdrives_mistral_2026,
author = {{AgentDrives}},
title = {Mistral AI: AI agent test drive},
year = {2026},
month = nov,
url = {https://agentdrives.com/tools/mistral},
note = {Agent test drive, CC BY 4.0},
}Something wrong on this page?
We check it against the drive and correct the page, with the date.