MCP Server

ThreadHound exposes a Model Context Protocol (MCP) server so you can connect an LLM client — Claude Code, Claude Desktop, claude.ai, Cursor, or your own agent — directly to your community data. Once connected, the model can list and search conversations, read messages, look up members, run analytics reports, and manage knowledge-base documents on your behalf.

Connecting to Claude (no API key required)

Claude Desktop / claude.ai (sign in with your browser)

These clients support a one-click Connect button — no API key needed. You sign in to ThreadHound in your browser and approve the connection.

Go to Settings → Connectors → Add custom connector.
URL: https://threadhound.ai/api/mcp/mcp/
Leave authentication set to the default and click Add.
In the next window, click Connect. A browser window opens ThreadHound's sign-in page.
Sign in with your ThreadHound account and approve access. The connector is now linked to your account — it can access every community you belong to, with your role in each.

Be sure to include the trailing slash '/' character when entering the URL

You might need to login twice when you do this - first login to Claude in your browser, then login to ThreadHound.

What can I do with the MCP connector

Connecting to other clients (API key required)

Get an API Key

Log in to the ThreadHound web app.
Go to Profile → API Keys → Create key.
Copy the key. Keys look like thk_ followed by 32 hex characters.

The key acts as you: it can access every community you are a member of, with your role in each. Treat it like a password — if it leaks, revoke it from the same API Keys page.

Connect Your Client

The URL is https://threadhound.ai/api/mcp/mcp/ — with a trailing slash.

Claude Code (CLI / IDE)

Terminal

claude mcp add --transport http threadhound \
  https://threadhound.ai/api/mcp/mcp/ \
  --header "Authorization: Bearer thk_YOUR_KEY_HERE"

Cursor and other JSON-configured clients

Add to your MCP config (e.g. .cursor/mcp.json):

.cursor/mcp.json

{
  "mcpServers": {
    "threadhound": {
      "type": "http",
      "url": "https://threadhound.ai/api/mcp/mcp/",
      "headers": {
        "Authorization": "Bearer thk_YOUR_KEY_HERE"
      }
    }
  }
}

Anthropic API (MCP connector)

Building your own agent? Pass the server directly in a Messages API call:

Messages API request body (excerpt)

{
  "mcp_servers": [
    {
      "type": "url",
      "url": "https://threadhound.ai/api/mcp/mcp/",
      "name": "threadhound",
      "authorization_token": "thk_YOUR_KEY_HERE"
    }
  ]
}

Testing with MCP Inspector

MCP Inspector is a handy way to explore the server interactively before wiring it into an agent:

Terminal

npx @modelcontextprotocol/inspector

Set the transport to Streamable HTTP, the URL to the endpoint above, and add the Authorization header. You can then list the available tools and invoke them interactively.

Asking Your First Questions

Once connected, you talk to your community data in plain English. Some prompts to try:

“List my communities.”

“Who posted about membership renewals in the last 3 months.”

“What analytics reports can I run?”

“Who are the top contributors?”

“Find negative, question-led conversations that never got a good answer.”

“Upload this FAQ document to the Policy category.”

💡 Good to Know

•Message truncation: message bodies are truncated to 1,000 characters, with a flag indicating when truncation happened.
•Privacy: member handles (e.g. phone numbers) are partially obfuscated, exactly as in the web UI.
•Errors are instructive: invalid category names, unknown ids, missing roles, and quota exhaustion come back as readable tool errors that tell the model (and you) what a valid call looks like.

AI credits: each semantic search consumes one AI credit from the community's monthly quota — exactly like the REST API's search endpoints. When the quota is exhausted the tool returns a clear error; retry with a non-semantic search type or contact your administrator.

Troubleshooting

Symptom	Likely cause
401 from the endpoint	Missing, invalid, or revoked API key — or wrong header format (must be `Authorization: Bearer thk_…`). For a browser-connected (OAuth) client this is normal on first connect and triggers the sign-in prompt; if it keeps re-prompting, your session expired — disconnect and reconnect the connector.
404 from the endpoint	Wrong path — it must be `/api/mcp/mcp/` including the trailing slash.
Tool error “not a member of this community”	The community id doesn't belong to this key's user — ask your agent to list your communities to see which ones the key can access.
Tool error about roles	Analytics reports and document upload/delete need an admin or owner role in that community.
Semantic search returns nothing	The community's AI credit quota may be exhausted — retry with a non-semantic search type, or check the response's notes for which retrievers actually ran.

Next Steps

🔐 Authentication

Learn how to create, manage, and revoke your ThreadHound API keys.

View authentication

🚀 REST API

Prefer direct HTTP calls? The same data is available through the REST API.

Getting started