5 min read
Nov 6, 2026

Use Pivot from your AI agent

Connect Claude Code, Claude Desktop, Cursor, or any MCP client to Pivot so your agent can read spaces and rooms and send messages.

Pivot runs a Model Context Protocol (MCP) server at https://api.pivot.app/mcp. Point an MCP-capable agent at it and the agent can list your spaces and rooms, read messages, and send messages, using the same API and the same permissions as the REST API.

Tool names are snake_case versions of the REST operations, for example get_room_messages and send_message.

Before you start

You need an API key, so keep it somewhere safe. Either type works: a personal API key makes the agent act as you, and an organization key makes it act as the integration. See API key types for how they differ.

The key goes in the Authorization: Bearer header. Replace YOUR_PIVOT_API_KEY in the examples below.

Claude Code

Terminal window
claude mcp add --transport http pivot https://api.pivot.app/mcp \
  --header "Authorization: Bearer YOUR_PIVOT_API_KEY"

Claude Desktop

Add Pivot to claude_desktop_config.json. Claude Desktop starts local servers, so use the mcp-remote bridge to reach the hosted one:

{
  "mcpServers": {
    "pivot": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.pivot.app/mcp",
        "--header",
        "Authorization: Bearer ${PIVOT_API_KEY}"
      ],
      "env": { "PIVOT_API_KEY": "YOUR_PIVOT_API_KEY" }
    }
  }
}

Cursor

Add Pivot to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "pivot": {
      "url": "https://api.pivot.app/mcp",
      "headers": { "Authorization": "Bearer YOUR_PIVOT_API_KEY" }
    }
  }
}

What the agent can do

Ask the agent to list the tools; it will show them. Reads and writes are both available. Tools that only read are marked read-only, and tools that delete are marked destructive, so most clients ask before running them.

  • List responses return 50 items by default. An agent can ask for up to 500 with the limit argument; larger requests are reduced to 500.
  • A tool is only offered if your key type can use that endpoint. A personal key sees the get_my_* tools; an organization key does not.
  • Errors come back with a code the agent can act on: auth, not_found, validation, rate_limited, or internal.

Privacy

Tool results pass through your agent host. Anything the agent reads from Pivot, such as messages, member names, and room details, is sent to the model provider behind that agent, under that provider’s terms. Only connect agents you are comfortable receiving that content, and prefer a key that can reach only the spaces the agent needs.

Limits in this version

There are no granular scopes yet. A key can do everything its owner or integration can do through the REST API, including sending messages and deleting content, and the agent can call any tool your key type allows. Revoke the key from your account settings if you stop trusting an agent.

Rate limits are the same as for the REST API, and each tool call counts as one request.

Was this guide helpful?

Prev

Message retrieval

Next

Webhooks