An MCP server for your requirements
Your agent can already read a file. What an MCP server adds is a spec it can query, update and not overwrite. Here is what exists, and what each one actually serves.
Where this comes from
Written by the Entalpa team; Entalpa is one of the servers below. The other servers are described from their own repositories as of October 2026, with star counts and last-commit dates from GitHub on 29 October 2026. Spec Kit's MCP status is from its documentation, as covered on Entalpa vs Spec Kit. We have not measured how much any of these changes an agent's error rate.
Why put requirements behind MCP at all
The default way to give a coding agent requirements is a markdown file: a PRD in specs/, referenced from CLAUDE.md or AGENTS.md. For a small project it is the right answer. It runs out in three places: the file is in a directory the agent can write to, so "do not edit" is a request; the agent reads the whole file or none of it; and nothing tells it which requirement a change touches.
An MCP server turns the spec into tools the agent calls: list the requirements for this story, read this one, mark it implemented, ask an open question. The agent gets the requirement it needs instead of the whole document, and the server, not the agent, decides what can be changed.
What to look for
Structured objects rather than one blob of text, write access with limits the agent cannot lift, a status per requirement, and the same spec served to every client you use.
The options
| Server | What it serves | Agent can change | Notes |
|---|---|---|---|
| Entalpa | Stakeholders, stories and requirements with links, status and open questions | Unlocked requirements only | Hosted, OAuth sign-in, 16 clients documented; imports Spec Kit and OpenSpec |
| Task Master | Tasks parsed from a PRD, with dependencies and priorities | Tasks | Open source; the requirements stay in the PRD file |
| mcp-server-spec-driven-development | Prompts that generate requirements (EARS), a design and code, in order | Writes files | Open source, ~440 stars, last commit September 2025 |
| spec-coding-mcp (kevinlin) | A Kiro-style requirements → design → tasks workflow for other IDEs | Writes files | Open source, ~20 stars |
| Spec Kit's MCP server | Workflow only | — | Experimental, and serves no spec artifacts |
| Filesystem MCP on specs/ | Your markdown files | Anything in the folder | No structure or locking; the do-it-yourself baseline |
| Tessl, Context7 | Specs and docs for external libraries | — | Different problem: other people's APIs, not your rules |
The prompt-based servers generate spec files and step out of the way; they do not hold the spec afterwards. That is the main line between them and a store.
Connecting Entalpa
Entalpa runs as a remote server over Streamable HTTP with OAuth: no API key to paste, your client opens the sign-in on first connect. In Claude Code it is one command:
claude mcp add --transport http entalpa https://api.entalpa.com/mcpCursor, Codex, Windsurf, VS Code with Copilot, Gemini CLI, Kiro and the rest each have their own page under connect your agent. The Agent Skills, the Claude Code and Codex plugins and the connection settings for each client are open source in entalpa/entalpa-mcp. Reading and writing the spec over MCP does not spend credits.
Which one fits
- One person, one repo, a few rules: skip the server. A PRD in
specs/and a line inCLAUDE.mdpointing at it; the PRD template is a good start. - You need a PRD turned into tasks: Task Master.
- You want an agent to draft a spec, Kiro-style, outside Kiro: one of the prompt servers.
- The spec has to outlive the session, be shared by several agents, and not be rewritten by them: Entalpa.
For the wider landscape, including the tools that are not MCP servers, see the best spec-driven development tools.
Questions this raises
Is there an MCP server for requirements and specs?
Yes. Entalpa serves structured requirements (stakeholders, stories, requirements with status and locking) over MCP to Claude Code, Cursor, Codex and other clients. Task Master serves tasks parsed from a PRD, and several open-source servers generate spec files from prompts.
Why not just point the filesystem MCP server at my specs folder?
It works, and it is a fine start. You get your files, not structure: no status per requirement, no links between them, and nothing that stops the agent editing a requirement you meant to be fixed.
Does an MCP requirements server replace CLAUDE.md or AGENTS.md?
No. Keep build commands, conventions and layout in the file. The server holds the domain rules, the part that outgrows a file first.
Entalpa gives your agent one locked source of truth
Structured, traceable requirements your agent reads over MCP — so it works from a curated spec instead of reconstructing your project every session. Free to start, no card, and MCP calls never spend credits.
Other guides