How it works
1
Connect MCP to your agent
Add the authenticated TRM API Docs MCP server to Claude, Cursor, Codex, VS Code, or another compatible agent.
2
Authenticate
Your agent opens the TRM API Docs sign-in page in a browser. Sign in, then follow the redirect back to your agent.
3
Retrieve current guidance
For your prompt, the agent reads a matching skill resource and retrieves the minimum current API pages and guides.
4
Build in your workspace
Your coding agent uses the retrieved contracts and skill instructions to generate or modify code, add tests, and report assumptions or missing choices.
Connect your agent
Use the authenticated MCP endpoint:Authentication is client-specific. Follow the authentication step in your client tab below to open the TRM API Docs sign-in flow in your browser.
- Claude Code
- Claude
- Cursor
- Codex
- VS Code
- Other MCP clients
Add TRM API Docs at user scope so it is available across projects:To authenticate from within Claude Code, start Claude Code and enter:Select Either method opens the TRM API Docs sign-in flow in your browser.Confirm that the server is configured:See the Claude Code MCP documentation for client-specific options.
trm-api-docs, then select Authenticate.Alternatively, trigger authentication directly from your terminal:claude mcp login must run in an interactive terminal. In a non-interactive or CI shell it fails with stdin isn't a terminal — authenticate once interactively (or use /mcp inside Claude Code), then reuse the stored credentials.Verify the connection
Start a new agent conversation and ask:UseA successful connection should show that your agent:trm-api-docs, not model memory. Find and read the Compliance API authentication guide (/guides/public/Authentication). Return its title and URL and confirm it lists both API Key and OAuth2 (Recommended). Then list the TRM MCP resources available in this session.
- Uses the documentation search capability to find the relevant page.
- Retrieves the full page through the TRM API Docs filesystem capability.
- Cites or links to the TRM API Docs page it used.
- Lists any TRM workflow skills available to your account when the client supports MCP resources.
Start building with TRM
You can use the connection for a focused API question, a single implementation, or a complete workflow.Connecting the MCP uses browser sign-in and grants your agent access to TRM documentation only. The integration code your agent writes authenticates to the TRM API separately — with your own API key or OAuth client. Generate one at my.trmlabs.com → Settings → API Tokens and keep it in an environment variable or secret manager, never in the prompt or the code.
Answer an API question
Using trm-api-docs, find and read the current documentation for the Wallet Screening endpoint that submits one or more addresses for screening. Explain the request, response, authentication, errors, and operational constraints. Cite every TRM API Docs page you use.
Integrate with a TRM API
Using trm-api-docs, implement a Wallet Screening client in this repository for one address and chain. Read the current endpoint, authentication, and error-handling documentation before editing code. Follow the repository’s existing conventions, add tests, and report the documentation pages and assumptions used.
Build a common compliance workflow
Using trm-api-docs, discover and read the TRM Compliance API integration skill. Help me implement a withdrawal workflow. Ask for any choices that materially change the implementation before editing code.
Troubleshooting
The browser does not open
Confirm that your client supports OAuth for remote HTTP MCP servers. Remove and re-add the server, or use the client’s manual authentication command.Authentication succeeds but protected pages are missing
Open the page directly in TRM API Docs with the same account. If you cannot access it there, request the appropriate API documentation permission and authenticate again.Tools work in the session but the server is not listed in /mcp
The server can be authenticated and its tools usable in the current session yet not appear in the persisted /mcp list. Reload the client, or remove and re-add the server, to persist it. This is a display or persistence quirk, not an authentication failure.