MCP Server for AI Assistants

New

Connect Claude Desktop, Cursor, and any MCP-compatible assistant directly to your BigQuery analytics. Read campaigns, explore audiences, and run workflows using real data natively in your chat.

JSON-RPC 2.0Streamable HTTPEnterprise TierTools & Prompts

Why connect your AI assistant

The Model Context Protocol (MCP) lets your AI assistant read your Mailchimp analytics natively. Ask complex questions and get answers grounded in real metrics from your BigQuery pipeline.

Real data, not guesses

Answers are grounded in actual Mailchimp data extracted to your BigQuery warehouse, eliminating AI hallucinations.

Ask in plain language

No need to write SQL or assemble complex REST payloads. Just ask your assistant to find your most engaged contacts.

Ready-made sales workflows

Generate call sheets, post-send strike lists, and pre-call dossiers effortlessly using built-in MCP prompts.

Whole-audience reach

Analyze campaigns, perform A/B variate testing, isolate at-risk subscribers, and view demographic distributions directly.

Secure OAuth 2.1 authentication

Your AI client connects via a secure browser-based sign-in flow. No API keys to copy, manage, or rotate. Revoke access anytime from your account settings.

Data-freshness visibility

Assistants can securely read recent data-pipeline sync statuses to guarantee the numbers they are reading are fully up to date.

Works with your tools

MCP InspectorClaude DesktopCursorVS CodeAny HTTP MCP Client

Works with any client that supports remote MCP servers over Streamable HTTP (SSE). For local clients that do not natively support browser-based OAuth flows, we fully support connection via the standard mcp-remote proxy.

What your assistant can do

Tools

These are the actions an assistant can call directly.

Prompts

These are ready-made workflows a user can invoke by name in their AI client. Support for prompts is still patchy across clients; you can expand any prompt to copy its full text and paste it into your assistant manually. Signed-in users can pick real values from their project to build a ready-to-run prompt.

Resources

These are reference documents an assistant can read for context.

Connect in under two minutes

Create an account on the Enterprise plan

MCP access requires an Enterprise subscription. You can create a free account first, or compare plans.

Point your AI client at the server

Choose one of the following methods based on your client's capabilities (do either 2a or 2b, not both):

Paste the server URL

For clients with a UI for adding MCP servers (like the official MCP Inspector), add this URL:

https://spacemonkey.igods.com/_api/public/v1/mcp

Your AI client will handle OAuth authorization automatically — it will open a browser window to sign in to Space Monkey. No API keys to copy or paste.

Or edit your config file

For clients that cannot perform OAuth automatically (like Claude Desktop or Cursor), we support connection via the standard mcp-remote proxy. Merge this snippet into your claude_desktop_config.json:

{
  "mcpServers": {
    "space-monkey": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://spacemonkey.igods.com/_api/public/v1/mcp"
      ]
    }
  }
}

Verify connection

Ask your assistant to call sm_list_projects to verify the connection is successful and map your company name to a Project ID. If you want to manually verify the server is operational without credentials, visit our unauthenticated health endpoint.

Note: The Space Monkey MCP server is strictly stateless. Tool discovery (via tools/list, prompts/list, and resources/list) does not require credentials, allowing tools like the MCP Inspector to load schemas before you sign in. However, actually executing a tool (tools/call) requires your active OAuth session.

Plans and limits

Access to the Space Monkey MCP server requires an Enterprise subscription. Connect your local AI tools and scale your automation using your standard Enterprise API keys.

Pro

Not included

—

Enterprise

Included

5,000 requests/hour · 500,000/month

Note: The MCP server executes actions seamlessly against your Enterprise API limits. A single tool call debits both your mcp usage quota and your underlying enterprise API quota. Multi-step AI workflows can deplete hourly quotas quickly.

How the server behaves

Project Context is required: The projectId argument is required on every tool and prompt except sm_list_projects. Your agent should always call sm_list_projects first to resolve a requested company name into its opaque projectId string.
Pagination formatting: List tools use opaque Keyset Pagination. Pass the cursor token from the previous response exactly as-is to fetch the next page. The pageSize is rigidly clamped to a maximum of 100 to prevent overwhelming LLM context windows.
Strict statelessness: The MCP endpoint is completely stateless and maintains no HTTP sessions. GET requests to the base endpoint yield HTTP 405 Method Not Allowed, and DELETE yields HTTP 404.
Read-only surface: The MCP gateway currently exposes purely read-only insight commands. There is no write-access surface exposed to AI assistants.

Troubleshooting common errors

HTTP 401 Unauthorized

The API key is missing, malformed, or invalid. Ensure you are sending it accurately as Authorization: Bearer sm_... in your proxy configuration or client setup.

HTTP 403 INSUFFICIENT_TIER

The key is valid, but the billing account it belongs to is not on the Enterprise tier. The MCP integration is exclusively an Enterprise entitlement.

HTTP 429 Too Many Requests

You have exceeded your account's rate limits. Check your API Keys dashboard to view your consumption windows or rotate keys as necessary.