apex-mcp-server

A read-only Model Context Protocol server for the Apex Rides API, deployed as a Cloudflare Worker. It authenticates against AWS Cognito with a service account and exposes four read-only tools over MCP's Streamable HTTP transport.

Tools

All tools are GET-based. There are no write operations and no generic passthrough.

Related MCP server: canvasMCP

Setup & Deploy

1. Install

npm install

2. Set secrets

npx wrangler secret put APEX_SERVICE_PASSWORD   # the Cognito service user password
npx wrangler secret put MCP_SHARED_SECRET       # a strong random string

For local development, create a .dev.vars file (git-ignored) with the same keys:

APEX_SERVICE_PASSWORD=...
MCP_SHARED_SECRET=...

3. Deploy

npx wrangler deploy

Usage

Connect an MCP client to the worker's MCP endpoint, with the shared secret as the first path segment:

https://<worker>.workers.dev/<MCP_SHARED_SECRET>/mcp

Any request whose first path segment is missing or does not match MCP_SHARED_SECRET gets a 401.

The transport is MCP Streamable HTTP (a single endpoint handles both the GET stream and POST messages). The legacy Node-only SSE transport does not run on Cloudflare Workers, so it is not used here.

Notes & Guardrails

• No client secret. SECRET_HASH is omitted entirely since the Cognito app client has no secret.

• Token caching. The access token is cached in module scope and refreshed via REFRESH_TOKEN_AUTH when near expiry.

• Date validation. popular_classes validates from/to against YYYY-MM-DD and appends T23:59:59Z to include the full final day.

• Name search. Since the Apex API has no text search, search_classes fetches up to 500 results and filters client-side when name is provided.

• Read-only enforcement. Only the four GET-based tools above are implemented.