anchor-mcp

anchor-mcp

An MCP server that reads a team's interviews, tickets, and docs to build a model of the product, exposing tools to ask questions, list contradictions and gaps, and draft documents, with all answers citing their source fragments.

Category
Visit Server

README

<img src="https://raw.githubusercontent.com/amitboker/anchor-mcp/main/.github/assets/banner.jpg" alt="Anchor MCP — ask your product's own history, get answers that cite their sources." width="100%">

CI MCP Node License

Anchor reads a team's interviews, tickets and docs and builds a model of the product — the decisions, the contradictions, and everything nobody has answered. This is that model, exposed to your agent over the Model Context Protocol.

Six tools. All six return the fragments an answer came from, because an answer you cannot check is a guess with better formatting.


Quickstart

Runs with no account, no key and no network, against a bundled sample project. Pick your client:

Claude Code

claude mcp add anchor -- npx -y github:amitboker/anchor-mcp

Claude Desktop — claude_desktop_config.json

{
  "mcpServers": {
    "anchor": {
      "command": "npx",
      "args": ["-y", "github:amitboker/anchor-mcp"]
    }
  }
}

Cursor — .cursor/mcp.json

{
  "mcpServers": {
    "anchor": {
      "command": "npx",
      "args": ["-y", "github:amitboker/anchor-mcp"]
    }
  }
}

VS Code

code --add-mcp '{"name":"anchor","command":"npx","args":["-y","github:amitboker/anchor-mcp"]}'

Restart the client afterwards — all of them read MCP config at boot.

Not on npm yet, so the specs above install from this repository. npx builds it on first run, which takes a few seconds once and is cached after that.


Tools

Tool What it does
ask_brain Ask a question about the product. Returns the fragments that bear on it, each with its source.
list_contradictions Statements that cannot both be true, both sides quoted, and the question that resolves each.
list_gaps Questions no source answers, who can answer them, and what they block.
list_sources Everything the brain has read, and how many fragments came out of each.
get_fragment Open one fragment by id — full text and where it sits in its source.
draft A PRD section, user stories or acceptance criteria, every line traced to a fragment.

What comes back

$ list_contradictions

3 contradictions in checkout-redesign.

## Cart retention window  (c-01)

A. The cart persists for 30 days on an unauthenticated session.
   — prd-checkout-v3 · §3.1 (PRD — Checkout redesign, v3)

B. Unauthenticated carts must expire within 7 days.
   — sec-review-q2 · Findings, item 4 (Security review — payments surface)

→ Does guest checkout keep a 30-day cart and require an identity, or drop
  to a 7-day window and stay anonymous?

Two documents, four months apart, in two different tools. Neither author was wrong. Nobody was going to notice.

Two things it will not do

It does not answer what it has not read:

$ ask_brain "quarterly revenue in Belgium"

Nothing in checkout-redesign answers that.

And it does not draft without sources:

$ draft --kind prd_section --topic "offline sync"

Not drafting "offline sync" — nothing in checkout-redesign supports it.

Both are the product working. A tool that fills a gap it cannot see the bottom of is the failure mode this exists to remove.


Sample data, and real projects

Out of the box the server runs on a bundled fixture — a checkout-redesign project with six sources and ten fragments. Every source, quote and person in it is invented, and every response says so in its footer.

Set ANCHOR_API_KEY and the same six tools read a real Project Brain instead. The output shape does not change; only the data does.

{
  "mcpServers": {
    "anchor": {
      "command": "npx",
      "args": ["-y", "github:amitboker/anchor-mcp"],
      "env": { "ANCHOR_API_KEY": "${ANCHOR_API_KEY}" }
    }
  }
}

Reference the variable from your shell rather than pasting a key into a file you might commit.

Status: preview. The tool surface is stable enough to build against. The hosted API behind ANCHOR_API_KEY is not open yet — until it is, every answer comes from the sample project.


Development

npm install
npm run build
npm run inspect

npm run inspect opens the MCP Inspector against your local build, which is the fastest way to see a tool's real output while changing it.

src/
  index.ts    server + tool registration
  tools.ts    tool bodies, callable without a transport
  brain.ts    the sample project
  cite.ts     citation formatting

One rule worth knowing before you add a tool: every response goes through answer() in cite.ts, and its cites argument is a non-empty tuple. An uncited answer is a type error rather than a matter of discipline — which is the only way a rule like that survives a deadline.

Logging goes to stderr, never stdout. stdout is the JSON-RPC transport, and one stray console.log on it corrupts the stream and takes the session with it.


Links

License

MIT — see LICENSE.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured