Anchor MCP Server

Decision Intelligence for Claude Code, Cursor, and MCP-compatible AI tools.

Anchor gives your AI coding assistant the ability to check decisions for bias, analyse trade-offs, record decisions into a system of record, and learn from your decision-making patterns over time.

Quick Start (2 minutes)

Prerequisites: Node.js 18+ and an Anchor API key.

1. Get an API Key

Ask your Anchor admin for a key, or generate one via the API:

curl -s -X POST https://ca-anchor-api-dev.orangesmoke-ddb1ea02.uksouth.azurecontainerapps.io/api/apikeys \
  -H "Authorization: Bearer <your-token>" \
  -H "Content-Type: application/json"

Save the returned apiKey value immediately — it is shown only once.

2. Configure Your AI Tool

Claude Code — Add to .mcp.json in your project root:

{
  "mcpServers": {
    "anchor": {
      "command": "npx",
      "args": ["-y", "@rubiconsoftware/anchor-mcp"],
      "env": {
        "ANCHOR_API_KEY": "ak_live_your_key_here",
        "ANCHOR_API_URL": "https://ca-anchor-api-dev.orangesmoke-ddb1ea02.uksouth.azurecontainerapps.io"
      }
    }
  }
}

Windows users: Use cmd /c npx instead of npx:

"command": "cmd",
"args": ["/c", "npx", "-y", "@rubiconsoftware/anchor-mcp"]

Cursor — Add to MCP settings in Cursor preferences with the same command/args/env.

Important: Add .mcp.json to your .gitignore — it contains your API key.

3. Add Decision Protocol to CLAUDE.md

Add these lines to your project's CLAUDE.md to tell the AI when to use the tools:

## Decision Protocol

Before committing to architecture choices, technology selections, design trade-offs,
effort estimates, or irreversible decisions:
1. Call `anchor_check_decision` with the situation — if CAUTION or STOP,
   call `anchor_analyze_decision` for deeper analysis.
2. After a decision crystallizes, call `anchor_capture_decision` to record it.
3. When uncertain about precedent, call `anchor_get_history` to check for
   similar past decisions.

4. Verify

Start a Claude Code session and ask it to make a decision. You should see it call anchor_check_decision.

Tools

| Tool | Description | Tokens | |---|---|---| | anchor_check_decision | Lightweight checkpoint — returns CLEAR/CAUTION/STOP with bias flags | ~150 | | anchor_analyze_decision | Deep analysis with bias detection, reframing, recommendations | ~300-500 | | anchor_capture_decision | Record a decision into the system of record | ~50 | | anchor_get_history | Search past decisions for precedent | ~100/entry | | anchor_get_profile | Your cognitive decision-making profile | ~200 |

Environment Variables

| Variable | Required | Default | Description | |---|---|---|---| | ANCHOR_API_KEY | Yes | — | API key starting with ak_live_ | | ANCHOR_API_URL | No | http://localhost:5000 | Anchor API base URL |

Development

If you're contributing to the MCP server itself:

cd mcp-server
npm install
npm run build
npm run dev    # Run with tsx for development

Architecture

This MCP server is a thin wrapper (~160 lines) over the Anchor REST API. All decision intelligence — bias detection, cognitive profiling, two-stage LLM analysis — lives in the API, not here.

See ADR-010 for design rationale.