Connect an agent with MCP

Connect a client, verify your workspace, discover tools, and handle metered calls.

Use Model Context Protocol (MCP) to connect an AI client to your GRID workspace. The server uses Streamable HTTP at https://app.startupgrid.ai/api/mcp.

Connect and check your workspace

  1. Open Settings → MCP & API. Follow the setup instructions for your client. For example, in Claude Code, run the command below, then open /mcp to sign in.

    claude mcp add --transport http grid https://app.startupgrid.ai/api/mcp

  2. When prompted, sign in and approve read-only access. If your client uses a bearer token, open API keys, enter a label, and select Generate Key. Paid Founder, VC, Ecosystem, and Enterprise plans include MCP keys. Save the key when shown and configure your client to send Authorization: Bearer <YOUR_GRID_API_KEY>. Keep it in your client’s secret settings. If exposed, select Revoke next to the key in GRID.

  3. Ask the client to call whoami and confirm the workspace and plan. Example prompt: “Use GRID’s whoami tool to show the connected workspace and plan. Then list the tools available to this connection.”

Discover tools and use their schemas

After the MCP handshake, call tools/list with your authenticated connection. Use the returned input schema for each tools/call. The server filters tools by role, plan, and authentication method. A first read-only request is:

{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"whoami","arguments":{}}}

Send it through the established MCP connection. Check the returned identity and plan, then choose a tool that fits the task.

Task Tools to look for
Check identity or research investors whoami, vc_directory_search
Review founder work get_pitch_score, list_matches, list_runs
Inspect investment work list_deals, get_deal, get_thesis
Inspect programme work list_events, list_applications
Inspect corporate work list_mandates, list_pilots

Metered tools

OAuth sign-in is read-only. An eligible API-key connection may expose metered tools. Confirm the workspace and the user’s approval for the cost before calling one.

  • dialectic_score_deck takes deck_markdown and returns a score and rationale. The metered cost is 50 credits.
  • legal_risk_check takes document_text and returns findings on the text. The metered cost is 20 credits. Check the response for unassessed areas.

If authentication fails, reconnect or check the key status. If a tool is missing, rerun tools/list and check your role, plan, and sign-in method. If GRID reports insufficient credits, check Billing before retrying. See pricing and credits for charge and refund details.