Firecrawl

Agents can search the web, scrape a page into clean text, and parse documents through Firecrawl. With an account connected, they can also run larger crawl and research jobs.

Firecrawl appears in the catalog on every instance; no experimental setting is needed.

Before you connect

Pick a method first, because they differ a lot in reach and cost:

Method What you need What agents get
Try keyless access Nothing Search, Scrape, and Parse only, with a limited free allowance
Sign in with Firecrawl A Firecrawl account and team The full account tool set, including jobs that spend credits
Use a Firecrawl API key An API key for the right Firecrawl team The full account tool set, for unattended or higher-limit use

For browser sign-in, ThinkingMach must be reached over HTTPS, or over a local http:// loopback address when it runs on your own machine.

Connect Firecrawl

  1. Open Connectors and select Firecrawl.
  2. Read the access line above the main button. It says who the connection belongs to and which agents can use it. Select Change to pick a different method, identity, or set of agents.
  3. Finish the method you chose:
    • Try keyless access — connect. There is nothing else to enter.
    • Sign in with Firecrawl — sign in and choose the Firecrawl team whose usage should apply. ThinkingMach registers its client automatically and asks for the firecrawl:global and offline_access scopes.
    • Use a Firecrawl API key — create a key for the intended team in Firecrawl, then paste it into the Your Firecrawl key field.

ThinkingMach stores the key as a secret and sends it to Firecrawl as a Bearer header; agents never see it. ThinkingMach cannot give a key more permissions than Firecrawl did.

Choose access

The keyless method exposes three read actions, firecrawl_search, firecrawl_scrape, and firecrawl_parse. It is the easiest place to start.

Sign-in and API key connections expose more. Crawls and research jobs run on Firecrawl's side and draw on your team's credits, and a broad crawl can use a lot of them quickly. New connections start with every action Allowed for all agents, so before you share an account connection:

  • Open the Permissions tab and set crawl and research jobs to Ask first.
  • Keep the connection on the team you are happy to bill.

See Set action permissions.

Try it

Use Firecrawl to scrape https://thinkingmach.com and give me the page title and a two-sentence summary. Do not start a crawl.

A single scrape is cheap and easy to check against the page in your browser. If you use the action tester on the Permissions tab instead, open More options, fill in only the URL, and leave the nested optional settings untouched.

Note: Illustrative task, not a recorded test result.

Troubleshooting and limitations

Problem Likely cause Fix
Keyless calls stop working The keyless allowance ran out Wait, or connect with a Firecrawl account or key
Only three actions are listed You connected with Try keyless access Connect again with sign-in or an API key for the full set
Credits drain faster than expected An agent started crawl or research jobs Set those actions to Ask first, and check job history in Firecrawl
The key is rejected It was revoked, or belongs to another team Create a new key for the right team and reconnect
Needs attention The grant or key was revoked Select Reconnect

Limitations: one Firecrawl team per connection. The available actions are whatever Firecrawl's hosted server exposes for the method you chose.

Sources