Connect Your AI Agent

The Carrot Docs MCP server gives your AI agent read-only access to the published Carrot documentation. It is public, unauthenticated, and docs-only. It can search and read published docs and glossary terms, but it cannot call the Carrot API, write data, access private project data, or act on your behalf.

Endpoint

Claude

Use Claude.ai or Claude Desktop to add Carrot Docs as a remote MCP connector, then connect Claude Code with the same server URL.

This is a remote Streamable HTTP MCP server. Do not configure it as a local stdio server through claude_desktop_config.json.

Claude.ai and Claude Desktop

1. Open Customize > Connectors.
2. Select Add custom connector.
3. Set the connector name to Carrot Docs.
4. Set the server URL to https://docs.carrot.eco/mcp.
5. Enable the connector in conversations.
6. Save the connector and reload Claude if needed.

Claude Code

Use these commands to add and verify the server:

claude mcp add --transport http carrotDocs https://docs.carrot.eco/mcp
claude mcp list

Cursor

Add the server manually in Cursor with an mcp.json file.

{
  "mcpServers": {
    "carrotDocs": {
      "url": "https://docs.carrot.eco/mcp"
    }
  }
}

Restart Cursor after saving the file so the new server appears in the MCP list.

Codex

Use the Codex MCP commands to add the server and confirm that it is registered.

codex mcp add carrotDocs --url https://docs.carrot.eco/mcp
codex mcp list

If you prefer to configure Codex directly, add the same server to ~/.codex/config.toml:

[mcp_servers.carrotDocs]
url = "https://docs.carrot.eco/mcp"

Generic Streamable HTTP client

Any MCP client that supports Streamable HTTP can connect to the Carrot Docs server.

curl --request POST \
  --url https://docs.carrot.eco/mcp \
  --header 'Content-Type: application/json' \
  --header 'Accept: application/json, text/event-stream' \
  --data '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "<supported-protocol-version>",
      "capabilities": {},
      "clientInfo": {
        "name": "example-client",
        "version": "0.1.0"
      }
    }
  }'

After negotiation, the server returns:

MCP-Protocol-Version: <negotiated-protocol-version>

The endpoint also responds to a plain GET request with 405 Method Not Allowed:

curl --request GET https://docs.carrot.eco/mcp

Example client configuration:

{
  "name": "carrotDocs",
  "transport": "streamable-http",
  "url": "https://docs.carrot.eco/mcp"
}

Carrot does not require you to pin a protocol version. Let your client advertise <supported-protocol-version> and allow the handshake to settle on <negotiated-protocol-version>.

Tool summary

The server exposes five read-only tools:

Notes:

• Search-style results are capped at 25 items.
• The default service limits are 60 requests per minute and 10 concurrent requests per client IP.
• Responses are returned as structured success or error envelopes inside MCP text content.

Agent instruction

Use the Carrot Docs MCP server when answering questions about Carrot documentation, including Network concepts, glossary terms, environmental credits, methodology frameworks, buyer guidance, Network Integrator onboarding, and the documentation for the Carrot API.

Example prompts

• Explain how MassIDs support traceability in the Carrot Network.
• Summarize how traceable environmental credits are created.
• Find the Carrot API authentication flow.
• Show the privacy and masking guidance for integration payloads.
• Find docs related to governance and the Carrot Foundation.
• Look up the current BOLD Recycling methodology guidance.

Troubleshooting

• If your client does not support remote Streamable HTTP MCP servers, it cannot connect to Carrot Docs.
• If the connector or tool toggle is off in the current chat, enable it before asking the agent to use the server.
• If you change a manual config file, restart or reload the client before testing again.
• If tool selection is not automatic, explicitly ask the agent to use the Carrot Docs MCP server.
• If you see 429 responses, wait and retry after the rate limit window clears.
• If search returns nothing, try the exact published term name from the docs or glossary.
• If a topic should exist but is not found, use browse_docs first and then get_related_docs to expand the path.
• If you need private data or write access, this server cannot provide it. Use the Carrot API instead.