Descriptions are omitted by default and the listings are paged, which is what keeps a full PRD set from swallowing the context window — pull the detail only for the document you land on. Drop a .specifai-path file in the project root and the path questions stop; without one, every call has to carry a cwd.
A read-only server over a Specifai project's generated specification documents, exposing them as tools any MCP client can call. Paged, with descriptions left out unless you ask for them.
- Each document family on its own tool: business requirements, product requirements, non-functional requirements, user interface requirements, business process documents and test cases — returned as ID and title, with the description off by default
- The PRD hierarchy walked in order: user stories under a PRD, tasks under a user story, and a single task in full
- list-all-tasks for a flat inventory across every PRD and user story at once, and get-task-by-id when you already know the ID
- search over all documents by text content or by PMO ID, with the document type returned alongside the match
- Linkage IDs carried through wherever they exist: Azure DevOps and Jira IDs and issue types, plus PMO IDs
- set-project-path to point the server at a different project when it cannot infer one
Node.js >= 16.0.0, or Bun >= 1.0.0 on the Bun runtime, and a project whose specification documents were generated by Specifai. Run it over stdio with npx --yes @presidio-dev/specifai-mcp-server@latest. Every tool takes a cwd and the server tries to infer the project directory from the environment; put a .specifai-path file containing the absolute path in your project root so you never have to pass it, or call set-project-path once. The author marks the server experimental — the tool surface is expected to change.
One command — npx -y @presidio-dev/specifai-mcp-server@latest