← Back to Discover

WritBase

An open-source, MCP-native task registry for fleets of AI agents. Tasks live in Postgres on Supabase, agents get scoped permissions, can delegate to each other, and every change is recorded. Apache-2.0.

MCP serversTask management

What it does

WritBase is a shared, persistent task registry for AI agents and the humans who supervise them. The README argues that agents need a record that survives between sessions, rather than in-memory state, and keeps tasks in Postgres instead of scattered across files and chat threads. Agents connect over the Model Context Protocol, so no custom integration is needed. Each agent gets its own key and permissions scoped to a project and department: read, create, update, assign, comment and archive. Worker agents can list, create and update tasks, with version checks to prevent silent overwrites. Manager agents can also manage keys, permissions, projects and departments, query an append-only audit log, register webhooks and find other agents by capability. Agents can assign tasks to each other, with depth limits and cycle detection. The backend is Supabase (Postgres plus an Edge Function), and an optional Next.js dashboard is included. Setup is done with an npx CLI that configures credentials, applies the schema and creates agent keys, or by cloning the repository. The README shows connection settings for Claude Code, Cursor and VS Code, and says Supabase Cloud or a self-hosted Supabase can be used.

Capabilities

  • Persistent task registry in Postgres, reachable through MCP
  • Per-agent keys with scoped permissions by project and department
  • Delegation between agents with depth limits and cycle detection
  • Append-only provenance log of who changed what, when and why
  • Optimistic concurrency, cursor pagination, rate limiting and request logging
  • npx CLI for setup, schema migration and key management

When to use it

  • Several AI agents need one shared, durable list of tasks
  • You want to limit each agent to the projects and actions it needs
  • You need an audit trail of what agents changed in a task list