# TinyFish MCP AI integration on Definable

> AI-powered browser automation and web interaction agent.

## What this connects

AI-powered browser automation and web interaction agent. Enables AI to browse, interact with websites, and perform web tasks.

Vendor: https://agent.tinyfish.ai/mcp

## Tools available

**8** tools available. First 8:

- `TINYFISH_MCP_CANCEL_RUN` — Cancel run — Cancels a running or pending automation run by its ID. Idempotent: if the run is already in a terminal state (COMPLETED, FAILED, CANCELLED), returns the current status without error. Only PENDING and RUNNING runs will be actively cancelled.
- `TINYFISH_MCP_DISCOVER_RUN` — Discover run — Discovers the currently running automation for this session. Returns runId if a run is active, or null if not yet started.
- `TINYFISH_MCP_GET_RUN` — Get run — Retrieves details of a specific automation run by its ID. Returns status, result, error, and other metadata. IMPORTANT: Call this after run_web_automation errors or times out — the run is likely still executing. If you do not have a run_id, use list_runs to find it. Also use this to poll for completion after run_web_automation_async. Runs typically take a few minutes. Wait 30-60 seconds between polls.
- `TINYFISH_MCP_GET_STEPS` — Get steps — Retrieves the steps taken during an automation run, including screenshots.
- `TINYFISH_MCP_LIST_RUNS` — List runs — Lists automation runs with optional filtering by status and pagination support. Returns an array of runs with their details. Use this to find a recently created run when you do not have its run_id (e.g., after run_web_automation timed out before returning a run_id). Filter by status=RUNNING to find active runs.
- `TINYFISH_MCP_POLL_STATUS` — Poll status — Polls the current status of an automation run. Returns status, step, progress, and other metadata.
- `TINYFISH_MCP_RUN_WEB_AUTOMATION` — Run web automation — Executes web automation given a URL and a Natural Language description of the automation goal to be performed. Automation is mostly focused around navigating complex web pages and extracting data from web pages. Include _meta.progressToken in your request to receive progress notifications. WARNING: This tool may take several minutes. If it errors or times out, the run is STILL EXECUTING on the server. DO NOT call this tool again or call run_web_automation_async — both would create a duplicate run. Instead: if you have a run_id from progress notifications, call get_run. If you have no run_id, call list_runs to find it. Only create a new run after confirming the previous one finished.
- `TINYFISH_MCP_RUN_WEB_AUTOMATION_ASYNC` — Run web automation async — Starts a NEW web automation run asynchronously and returns a run_id immediately without waiting for completion. WARNING: Do NOT use this to retry a failed run_web_automation call — that run is still executing. Use get_run or list_runs to check its status instead. Only use this tool for genuinely new tasks. Workflow: 1) Call this tool to get a run_id. 2) Wait 30-60 seconds. 3) Call get_run with the run_id to check status. 4) Repeat step 2-3 until status is COMPLETED, FAILED, or CANCELLED.

## Auth

Auth schemes: `DCR_OAUTH`.

## How agents use TinyFish MCP

Inside a Definable workflow, TinyFish MCP is one of the tools the **Distributor specialist** can call. Example coordination patterns:

- **Researcher → TinyFish MCP** — the Researcher (GPT-5.5) pulls context from TinyFish MCP (records, threads, documents), synthesises findings, and briefs the rest of the team.
- **Writer → Distributor → TinyFish MCP** — the Writer (Claude Opus 4.7) drafts copy in brand voice, the Verifier passes it, then the Distributor writes the result into TinyFish MCP (create record, post message, draft email).
- **Designer / Engineer → Distributor → TinyFish MCP** — the Designer ships an asset or the Engineer ships a code change, the Distributor delivers it via TinyFish MCP (attach file, open PR comment, post status).

The Verifier checks every TinyFish MCP call. On rate limit, schema drift, or auth refresh it self-heals and retries — the workflow completes without manual intervention.

## Categories

- model context protocol — https://definable.ai/apps/category/model-context-protocol/
- ai agents — https://definable.ai/apps/category/ai-agents/
- developer tools — https://definable.ai/apps/category/developer-tools/

## Related

- HTML page: https://definable.ai/apps/tinyfish_mcp/
- Same category (model context protocol): https://definable.ai/apps/category/model-context-protocol/
- All integrations: https://definable.ai/apps/
- Workflow (multi-agent loop): https://definable.ai/workflow/
- Apps llms.txt index: https://definable.ai/llms-apps.txt