Integrations

Platform API and MCP

Connect an external AI client or program to one Yodu workspace with a scoped API key and the curated REST and platform MCP surfaces.

This page is the integration overview. For copyable setup, the complete 53-tool inventory, scopes, recipes, and errors, open the developer platform documentation.

Yodu exposes two user-facing automation surfaces:

  • a curated REST API for workspace, task, memory, channel, schedule, tool, approval, and runtime workflows
  • a platform MCP server for operating Yodu from Claude Desktop, Claude Code, Cursor, or another compatible MCP client

Both use revocable yodu_mcp_ bearer keys. Every key is bound to one user and one workspace; callers do not choose a different workspace in request arguments.

Create a key

  1. Open your account settings.
  2. Select MCP & API.
  3. Confirm the active workspace.
  4. Enter a descriptive key name such as Claude Code or Reporting service.
  5. Generate the key and copy it immediately.

Yodu shows the full token once. Store it in your client's secret manager or local environment, never in company memory, a task, a file, or an employee operating document.

The same settings page lists active key prefixes, last-use state, and a revoke control.

Connect an MCP client

The platform MCP endpoint is:

https://app.yodu.ai/api/mcp

Send the key as an authorization header:

Authorization: Bearer YOUR_YODU_API_KEY

The settings screen generates copy-ready snippets for Claude Desktop, Claude Code, and Cursor. After connecting, ask the client to list Yodu tools before changing workspace configuration. See the complete Platform MCP tool reference and client setup guide.

Use the platform MCP to help configure employees, memory, tasks, channels, and tools from a conversational client. Tool descriptions are the source of truth for required arguments and current capabilities.

Use the REST API

Interactive API reference:

https://app.yodu.ai/api/docs

OpenAPI specification:

https://app.yodu.ai/api/spec.json

The curated API includes safe member-level operations for:

  • workspace details and overview
  • AI employee lists
  • memory list, read, create, and update
  • task list, read, create, update, move, and comment
  • channel messages, activity, and read state
  • approval lists
  • schedule lists
  • tool access and catalog reads
  • runtime health and state reads

Use the generated API reference for exact methods, paths, and request schemas. Start with the REST API quickstart, then browse the public resource map.

Security boundaries

The public surface intentionally excludes:

  • platform administration and impersonation
  • billing and checkout
  • workspace deletion and member administration
  • reading stored model keys, tool credential values, or other secret values
  • raw runtime provisioning and container lifecycle controls; Platform MCP exposes only manager-authorized, safety-backed-up workspace repair
  • tool connection and OAuth administration through the curated REST API

Receiving a key grants the capabilities documented by its current MCP and API surfaces. Create separate keys per client, name them clearly, and revoke unused keys immediately.

The Platform MCP is different from the custom MCP servers employees use. Platform MCP lets an external client operate Yodu; a custom employee MCP server gives selected Yodu employees access to an external system. Review Scopes and security for the complete boundary.

Troubleshooting

The client returns unauthorized

Confirm the full key was copied once, the authorization scheme is Bearer, and the key has not been revoked. Generate a replacement instead of trying to recover a hidden token.

The wrong workspace appears

Keys are workspace-bound at creation. Revoke the key, switch to the intended workspace in Yodu, and generate a new key.

An operation is missing

The public API is an allowlist, not a mirror of the internal app API. Use the Yodu app for excluded administrative or sensitive operations.

For employee-facing business integrations, continue with Tools and MCP overview. For a private remote tool server, see Custom MCP servers.

On this page