MCP & AI agents
RyTask ships a first-party MCP server with full parity — anything a person can do in the UI, an AI agent can do through a tool, under the same permissions.
RyTask includes a first-party Model Context Protocol server, so AI agents — Claude Code, Claude Desktop, or any MCP-capable client — can work in your RyTask the way a teammate does: capture and update tasks, manage projects and statuses, search, comment, read notifications, and administer membership and tokens.
The parity promise
The MCP server is not a curated subset. RyTask's rule is 100% parity: every service
capability the application exposes has a corresponding MCP tool. Today that is 49 tools,
and the number is enforced, not aspirational — a CI check (scripts/check-mcp-parity.ts)
fails the build if a capability ever ships without its tool. When RyTask grows a feature,
agents get it too, by construction.
Just as important, the tools are not a parallel implementation. Each tool calls the same service code its REST and UI counterparts call, under the same tenancy and the same role-based permission checks. There is no separate "agent path" that could drift or leak.
The full catalog — names, inputs, and outputs for all 49 tools — is in the MCP tools reference.
How access works, in one paragraph
An agent authenticates with a personal access token that you mint in the app and can scope down (for example, read-only). The token decides everything: which organization the agent operates in, and what it may do — its effective permission is the token's scopes intersected with your role, default-deny. Work captured by an agent is stamped with its source and shows an "Agent" badge in the UI, so provenance stays visible.
In this section
| Page | What it covers |
|---|---|
| Connecting a client | The HTTP and stdio transports, with copy-paste setup for Claude Code and Claude Desktop. |
| Agent access & tokens | Minting, scoping, and revoking personal access tokens. |
| Security model | What the token controls, how tenancy and permissions are enforced, and which tools deserve a second look before running. |