MotherDuck
motherduck.comMotherDuck is a cloud database service built on DuckDB.
Can an AI agent use MotherDuck? Yes, without signing up. Our agent got into MotherDuck on its own in 2 minutes and did all 10 of its core tasks.
From finding it to leaving
What an agent managed without a person, at each step.
- On its own
Find
Docs it can act on: llms.txt, an OpenAPI and an MCP server.
596 doc pages read
- On its own
Get in
Try first, claim later, MCP sign-in, sign-up in a browser.
2 minutes
- On its own
Use
Did: run queries and create databases and tables; build a chart (Dive) and embed it in a page.
10 of 10 core tasks
- Not reached
Pay
Its robots.txt closes the payment page.
- On its own
Leave
Nothing to cancel, deletes its account.
Getting in
4 of 5 ways in worked without a person.
The way that worked
Try first, claim later2 minutes
- An inboxNot needed
- A browserNot needed
- A personNot needed
- Agent on a serverAPI, MCP, CLIPartlyWithout an account (try first, claim later) it gets an API key, makes its first API call and does all of the core work, then stops at payment: paying takes a checkout page in a browser.
- Assistant in its own browserLike ChatGPT agentNot checkedSign-up wasn't tried in this drive.
- Signed agentNot checkedNot checked: its robots.txt asks crawlers to stay off the sign-up page.
Every way in
| Way in | What happened | Why |
|---|---|---|
| Try first, claim later | Worked | |
| Email and password | Worked | |
| CLI code login | Worked | |
| Sign-in for apps (OAuth, MCP) | Worked | |
| Sign-up by API | Not tried | The same as another way |
- Crawlersrobots.txt asks crawlers to stay off sign-up and the API. Nothing to fix. It keeps these pages out of search and doesn't stop agents signing up.
What an agent can do with MotherDuck
All 10 core tasks done, and 20 of 23 others.
Core tasks
- done there
| Task | Result | Through |
|---|---|---|
| Run a SQL query against a MotherDuck database | Done | MCP: done thereCLI: done there |
| Create a database | Done | MCP: done thereCLI: done there |
| List databases and read their schema | Done | MCP: done there |
| Create a table | Done | MCP: done thereCLI: done there |
| Generate an interactive visualization (Dive) from a natural-language question | Done | MCP: done thereCLI: done there |
| Embed a Dive in an external application | Done | API: done there |
| Build and deploy a scheduled Python data pipeline (Flight) | Done | MCP: done thereCLI: done there |
| Trigger a Flight run on demand | Done | MCP: done thereCLI: done there |
| Read Flight run history and logs | Done | MCP: done thereCLI: done there |
| Build a managed Iceberg/DuckLake lakehouse on cloud object storage | Done | MCP: done thereCLI: done there |
Reliability
70of 100
Core tasks that worked on all 3 runs
Handles mistakes
72of 100
Clear errors, retries without duplicates, honest limits
Surface tested
97of 100
Documented routes, tools and commands run and checked
Docs accuracy
99of 100
Documented operations that behaved as the docs say
Docs completeness
75of 100
How fully the docs describe each operation
What it can't do, and why
- Pay for a plan Buying is a card form in the web app; no billing API or agent payment.
- Connect an Iceberg catalog Needs an outside catalog and cloud credentials the account lacked.
- Keep a terminal-made account past 72 hours A person must claim it first.
20 of 23 other tasks done
- Query a table as of a past time or snapshot version (time travel)Done
- Create and manage a read-only data share of a databaseDone
- Scope a share to specific tables and mask or filter rows/columnsDone
- Restore a database from a snapshotDone
- Create a named long-lived backup snapshotDone
- Write a Guide capturing org metric definitions and query conventions for AI agentsDone
- Read and browse existing Guides by topicDone
- Update or edit a Guide's contentDone
- Ask a natural-language documentation question (ask_docs_question)Done
- Search the catalog for databases, tables, shares and GuidesDone
- Auto-fix a broken SQL query (FixIt)Done
- Visualize result-set columns instantly (Column Explorer)Done
- Run SQL AI functions (prompt, embedding) over dataDone
- Scale read workloads with read-scaling replicasDone
- View consolidated query history across the organizationDone
- Ingest data from cloud storage, databases, or SaaS sourcesDone
- Resize compute instance type/sizeDone
- Create a service account for programmatic accessDone
- Restrict a connection to read-only database accessDone
- Assign role-based access control to usersDone
- Set up AWS PrivateLink or IP allowlisting for network accessPaid plan
- Read Apache Iceberg tables via REST catalog: Needs an outside catalog and cloud credentials the account lacked.Blocked
- Invite an internal user to the organizationNot tried
- Purchase via AWS Marketplace private offerNot tried
The plans they need
- Set up AWS PrivateLink or IP allowlistingEnterprise
Paid-plan tasks don't count toward the level. The drive used the no-card trial (Business, 7 days) on the web account; Free Plan on the CLI account.
Surface audit: 194 documented operations, 187 run, 186 passed, 2 not as documented, 1 blocked (sandbox or plan limits), 3 not run for time, 5 left out to stay safe.
How an agent connects
MCP, API, OpenAPI, CLI and llms.txt found.
MCP server
All testedOfficial server. Remote server, 47 tools when driven (SQL, schema, Dives, Flights, Guides, docs Q&A) Sign‑in needed to see its tools.
41 tools across its docs
API
All testedREST admin API, 10 operations (users, tokens, ducklings, embed sessions); data work is SQL.
10 operations across its docs
OpenAPI
Yes3.0.3, 10 operations.
CLI
All testedMotherDuck CLI v1.5.6-2026-10-707: new, login (device flow), query, dive, flight, guide.
35 commands across its docs
SDK
94 tested101 methods across its docs
llms.txt
YesAgent payments
None foundThe checkout: Chose a plan; the checkout didn't open: Business, $250 a month; Checkout: Stripe Checkout; No CAPTCHA at checkout.
Docs
No gaps in what we readCovers getting started, auth, error codes and code examples.
12 of 12 documented ways to do the core work worked when tried.
motherduck.com/docs/getting-started
596 of 743 pages read
Tips for agent builders
- Install the MotherDuck CLI from https://install.motherduck.com and run 'motherduck new -o json' to get an account and token with no browser; delete it with DELETE /v1/users/<username> when done.
- On an existing account, use 'motherduck login --headless' then 'motherduck login --device-code=<code>' after a person confirms the code.
- Use the remote MCP server at https://api.motherduck.com/mcp with 'Authorization: Bearer <token>' (protocol 2025-11-25); call get_dive_guide or get_flight_guide before writing Dives or Flights.
- The MCP 'query' tool refuses CALL, SHOW ALL ROLES and EXPLAIN ANALYZE as not read-only: send them through query_rw.
- Find your username with GET /v1/users before calling /v1/users/{username}/... (the email's local part with '-' turned into '_').
- Create scoped tokens with POST /v1/users/{username}/tokens (ttl >= 300, token_type read_scaling for read-only).
How MotherDuck compares
In Data warehouses
1st of 2 Data warehouses products our agent got into on its own.
| Product | Level | Find to leave |
|---|---|---|
| L4 | Find: On its own, Get in: On its own, Use: On its own, Pay: Card checkout, Leave: On its own. | |
| L4 | Find: On its own, Get in: On its own, Use: On its own, Pay: Not reached, Leave: On its own. | |
| L1 | Find: On its own, Get in: Stopped, Use: Not reached, Pay: Not reached, Leave: Not reached. | |
| L1 | Find: On its own, Get in: Stopped, Use: Not reached, Pay: Not reached, Leave: Not reached. | |
| L1 | Find: On its own, Get in: Stopped, Use: Not reached, Pay: Not reached, Leave: Not reached. | |
| L1 | Find: On its own, Get in: Stopped, Use: Not reached, Pay: Not reached, Leave: Not reached. | |
| L1 | Find: On its own, Get in: Stopped, Use: Not reached, Pay: Not reached, Leave: Not reached. |
Sorted by level, then work done, then time to the first call. How we rank
For the team behind MotherDuck
Our agent didn't reach Pay. Prove MotherDuck is yours, then see where each issue is, read your safety findings and ask for a free drive after a fix. How to fix each one comes with the Audit.
5 things to fix for agents
Agent blockers: 1 foundAgent surfaces: 2 issues foundSafety review: 2 issues found
- FindClear
- Get inClear
- UseClear
- PayNot reached
- LeaveClear
On our standard
29 criteria: against other data warehouses products where we measured 5 or more, otherwise against every product we measured.
- Export before leavingThe best result All 6 others (data warehouses products) lower.
- Every way in works for an agentAhead of most 5 of 6 others (data warehouses products) lower, 1 the same.
- All validation problems at onceAhead of most 29 of 35 others (every product) lower, 6 the same.
- Structured resultsAhead of most 26 of 36 others (every product) lower, 10 the same.
- Results sized for a context windowBehind most 16 of 31 others (every product) higher, 11 the same, 4 lower.
- Spend an agent can see and capLevel with most 3 of 5 others (data warehouses products) the same, 2 higher.
- Docs an agent can navigateIn the middle 3 of 6 others (data warehouses products) the same, 3 higher.
- A request id on every answerBehind most 3 of 5 others (data warehouses products) higher, 2 the same.
All 29 criteria
- A path written for agentsAhead of most 4 of 6 others (data warehouses products) lower, 2 the same.
- A person can claim what the agent madeAhead of most 5 of 9 others (every product) lower, 4 the same.
- The best way in is built for agentsIn the middle 3 of 6 others (data warehouses products) lower, 3 the same.
- Prices an agent can readLevel with most 4 of 6 others (data warehouses products) the same, 2 lower.
- Sign-in for agents acting for a personLevel with most 4 of 6 others (data warehouses products) the same, 2 lower.
- One way to page through listsIn the middle 35 of 87 others (every product) lower, 42 the same, 10 higher.
- Answers sized for an agentLevel with most 61 of 85 others (every product) the same, 24 lower.
- Limits known before they hitAhead of most 27 of 47 others (every product) lower, 5 the same, 15 higher.
- Fast to the first callLevel with most 3 of 4 others (data warehouses products) the same, 1 lower.
- Versions an agent can pinLevel with most 3 of 4 others (data warehouses products) the same, 1 lower.
- Descriptions that say when to use the toolIn the middle 20 of 65 others (every product) lower, 30 the same, 15 higher.
- Tool names that sort themselvesLevel with most 61 of 65 others (every product) the same, 4 lower.
- Real status codesThe same as all 5 others (data warehouses products).
- Errors a program can readIn the middle 1 of 4 others (data warehouses products) lower, 2 the same, 1 higher.
- Errors say what to do nextIn the middle 1 of 4 others (data warehouses products) lower, 2 the same, 1 higher.
- WebMCP tools on the web appThe same as all 5 others (data warehouses products).
- Typed, tight input schemasLevel with most 39 of 65 others (every product) the same, 10 lower, 16 higher.
- Pages fit one fetchLevel with most 5 of 6 others (data warehouses products) the same, 1 higher.
- Truthful annotationsBehind most 34 of 65 others (every product) higher, 13 the same, 18 lower.
- A tool list that fitsBehind most 33 of 65 others (every product) higher, 16 the same, 16 lower.
- Agents pick the right toolLevel with most 44 of 65 others (every product) the same, 1 lower, 20 higher.
Where MotherDuck fits
- Category
- Tasks
- Wall
- Card before the first useour agent got in another way
- Rivals
Found an error? Tell us on this page or read how corrections work.
Questions people ask about MotherDuck
- Can an AI agent sign up for MotherDuck on its own?
- Our agent got into MotherDuck without signing up. Its way in: Try first, claim later, in 2 min. Not needed on the main way: the documented CLI command
motherduck newmade an unclaimed Free Plan organization and token in 3 s with no form, email or browser; a person must claim it within 72 h to keep it. The web sign-up (email + password on auth.motherduck.com, no CAPTCHA) also worked in the browser; it asked for a name, a team name and a region; our agent answered. - Does MotherDuck have an MCP server, and can an agent connect to it?
- Yes. MotherDuck has an official MCP server. The server is hosted and lists 47 tools. A new agent can get in on its own: in the drive of 26 Oct 2026, our agent got its key without signing up, and without a person. Through the MCP server, 24 of our agent's 25 attempts worked.
- How does an AI agent get a MotherDuck API key?
- In the drive of 26 Oct 2026, our agent got its MotherDuck API key over the API, without a browser or a person. motherduck new saves a read/write token to credentials.json.
- Can an AI agent delete its MotherDuck account?
- Yes, over the API, without a browser. REST answered 404 at once; MCP and CLI refused the token from about 5 minutes later. The sign-in identity at auth.motherduck.com remains and only offers to create a new account.
- Can an AI agent pay for MotherDuck?
- No payment made: its robots.txt closes the payment page. No card needed to start.
Cite this page
The page as a reference, in plain text, APA and BibTeX.
Cite this
AgentDrives, “MotherDuck: AI agent test drive”, 26 October 2026, https://agentdrives.com/tools/motherduck. CC BY 4.0.
AgentDrives. (2026, October 26). MotherDuck: AI agent test drive. https://agentdrives.com/tools/motherduck
@misc{agentdrives_motherduck_2026,
author = {{AgentDrives}},
title = {MotherDuck: AI agent test drive},
year = {2026},
month = oct,
url = {https://agentdrives.com/tools/motherduck},
note = {Agent test drive, CC BY 4.0},
}Something wrong on this page?
Tell us what's wrong. We check it against the drive and correct the page, with the date.