Piyaz MCP
A self-hosted MCP server and web app that keep a project as a task graph shared by people and agents.
- 200 ★ on the repository
- Last commit
- AGPL-3.0
Broad access
Why this level
- Composer writes code, opens pull requests and fixes review notes without your involvement.
- The server keeps the project plan, decisions and work history in your PostgreSQL database.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/piyazDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Composer writes code, opens pull requests and fixes review notes without your involvement.
- The server keeps the project plan, decisions and work history in your PostgreSQL database.
This is third-party code. Review the repository files before installing.
What it does
Piyaz is a project management workspace where people and coding agents work on one project. The project is a graph: each task shows what blocks it, the decisions behind it and what is already done. Piyaz splits an idea or an existing codebase into small tasks, and an agent picks them up with the plan, standards and context. You run it yourself: the server works as a remote MCP at `/api/mcp` (Streamable HTTP, OAuth 2.1, no API keys), and nine tools cover the path from planning to review. Claude Code, Codex, Cursor and Antigravity share one view of tasks. Agent work stops at `in_review`, and a reviewer reads the diff and returns a verdict on security, performance and standards. In Claude Code the `/piyaz:composer` command carries a task from research to an open pull request on its own.
Who it is for. Teams where several agents and people run one project and need a shared plan with review before merge.
Good fit when
- An idea must be split into tasks with dependencies so agents do not collide.
- A new agent session must continue where the last one stopped.
- Agent work needs review with a file and line for every finding.
- You need one view of the project for Claude Code, Codex and Cursor.
Not a fit when
- You want a cloud solution with no server of your own: Piyaz runs on your own Bun and PostgreSQL.
- The AGPL-3.0 license is unacceptable for your product.
- The project is small and a task graph only adds overhead.
Example request
/piyaz I want a CLI that generates changelogs from conventional commits, with monorepo support.MCP
- Transport
- http
- Authentication
- OAuth
Limitations
You install your own server from source: Bun and Docker for PostgreSQL are required, and step-by-step instructions live at docs.piyaz.ai. The AGPL-3.0 license places conditions on network distribution of changes. The `/piyaz:composer` command works only in Claude Code and writes code, opens pull requests and fixes review notes by itself, so merging stays with you. The MCP server uses OAuth rather than keys and the client must support it. The last commit dates to October 2026.
How to disable. Remove the Piyaz entry from the MCP client config, stop your server and drop the PostgreSQL database if needed.
FAQ
What does it take to run?
Bun and Docker for PostgreSQL. See the self-host guide at docs.piyaz.ai for details.
How do I connect from an agent?
Register a remote server in the MCP client at `/api/mcp` of your instance. Authorization uses OAuth 2.1 and no API keys.
What does /piyaz:composer do?
In Claude Code it takes a task through research, plan, implementation, pull request and a bounded review fix loop, then picks the next ready task.
What does the in_review status mean?
Agent work never reaches `done` on its own: a review agent reads it and returns a verdict with a file and line for each finding.
Which agents work together?
Claude Code, Codex, Cursor and Antigravity see one task map.
Discussion
No questions or reviews yet.
Related
MCP serversA scaffold for Claude Code and Codex: tasks in files, developer and QA roles, manual approval and shared state.
A self-hosted issue tracker and wiki on Cloudflare with MCP: 126 tools, issue leases and file claims for several agents
Plane MCP Server: project management in Plane
Plane MCP Server
Official Plane MCP server: the agent manages projects, work items, cycles and releases through 30 tools
The MCP endpoint of a Salt workspace on your server: an agent creates pages, databases and comments with a person rights.
Paca is a self-hosted task tracker where AI agents work as Scrum team members alongside people
The official Taskade MCP and plugin: an agent manages workspace projects, tasks, automations and AI agents.
README in short
The README explains the workspace idea and shows the plan, pick, review and Composer loop with sample commands. A self-hosting section, the MCP server description and the stack follow.