MCP overview

Connect the Onetap dashboard to Claude and other AI clients, and work with your data by describing what you want.

Talk to your data

Point an MCP client at the endpoint and ask for what you want. No SDK and no export step — the server runs inside the app and answers as the person who signed in.

Connect a client Tool reference

tools/call

1{2  "jsonrpc": "2.0",3  "id": 1,4  "method": "tools/call",5  "params": {6    "name": "list_pipelines",7    "arguments": { "status": "active" }8  }9}

What you can build

Onetap ships a remote Model Context Protocol server as part of the app itself — there is no separate process to run. Today it exposes pipelines: projects made of ordered stages, the same objects the dashboard shows.

Create a whole project from a description — stages, goals, priorities and due dates in one call.
Read back what exists, including every stage and its status.
Edit deadlines, statuses and copy, with the change attributed to the person who made it.

This is the one developer surface that is live today. The attendance, presence and webhook surfaces are still proposed designs.

Endpoint

Everything runs through a single URL. Point any MCP client that supports remote Streamable-HTTP servers at it — see Quickstart for the exact steps.

Endpoint

https://onetaplabs.com/api/mcp

Protocol

The transport is JSON-RPC 2.0 over Streamable HTTP. Four methods are implemented:

initialize     # handshake, returns server capabilities
tools/list     # the five available tools and their schemas
tools/call     # invoke one tool
ping           # liveness check

Requests are stateless — there is no server-to-client streaming, and GET /api/mcp returns 405.

Using it for analytics

Because the client is an AI assistant, the useful pattern is asking questions rather than writing queries. Connect the server, then ask things like:

"Which pipelines have stages overdue by more than a week?"
"Summarise what moved to done across every project this month."
"Which stages have no due date set?"

The model calls list_pipelines and get_pipeline to gather what it needs, then answers. Nothing is exported and no credentials leave the client.

Access is still scoped by project membership, so a teacher connecting their own account sees only what they could see in the dashboard.