@glidepaths/mcp-server

MCP server for Glidepaths AI governance. Connects any Claude-powered agent or MCP-compatible framework to Glidepaths natively — log decisions, enforce policy guardrails, query compliance posture, and generate audit packages as first-class tools.

Tools

| Tool | What it does | |---|---| | log_decision | Writes an agent decision to the 14-field Glidepaths audit trail | | evaluate_action | Checks a proposed action against active governance policies before execution | | get_compliance_matrix | Returns control-by-control compliance posture (compliant / partial / gap) | | generate_evidence_package | Produces an audit-ready evidence bundle for a date range and framework |

Prerequisites

1. A Glidepaths account at glidepaths.com
2. An API key from Settings → API Keys
3. Node.js 18 or later

Installation

npm install -g @glidepaths/mcp-server

Or run without installing:

npx @glidepaths/mcp-server

Claude Desktop configuration

Add the following to your Claude Desktop config file.

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "glidepaths": {
      "command": "npx",
      "args": ["@glidepaths/mcp-server"],
      "env": {
        "GLIDEPATHS_API_KEY": "glide_your_api_key_here"
      }
    }
  }
}

Restart Claude Desktop after saving. You should see a hammer icon (🔨) indicating the tools are loaded.

Custom base URL (self-hosted / staging)

{
  "mcpServers": {
    "glidepaths": {
      "command": "npx",
      "args": ["@glidepaths/mcp-server"],
      "env": {
        "GLIDEPATHS_API_KEY": "glide_your_api_key_here",
        "GLIDEPATHS_BASE_URL": "https://staging.glidepaths.com"
      }
    }
  }
}

Environment variables

| Variable | Required | Default | Description | |---|---|---|---| | GLIDEPATHS_API_KEY | Yes | — | API key from Glidepaths Settings | | GLIDEPATHS_BASE_URL | No | https://glidepaths.com | Override for self-hosted or staging deployments |

Usage examples

Once connected, Claude can use the tools naturally in conversation:

User: Check our EU AI Act compliance posture. Claude: (calls get_compliance_matrix with framework: "EU_AI_ACT") Here's the current status...

User: Log that the claims triage agent just approved claim #4821 at medium risk. Claude: (calls log_decision with the relevant fields) Decision logged.

User: Before the agent sends this payment, check if it's allowed. Claude: (calls evaluate_action) The policy engine returned proceed — all active policies passed.

User: Generate a Q1 2025 audit package for NAIC. Claude: (calls generate_evidence_package) Package generated covering Jan 1 – Mar 31, 2025.

Using with other MCP frameworks

The server speaks standard MCP over stdio and works with any MCP-compatible host:

# Run directly
GLIDEPATHS_API_KEY=glide_xxx node dist/index.js

# Or with npx
GLIDEPATHS_API_KEY=glide_xxx npx @glidepaths/mcp-server

Building from source

cd sdk/mcp
npm install
npm run build     # compiles TypeScript to dist/
npm start         # runs the compiled server

Tool reference

log_decision

{
  "agent_id": "claims-triage-v2",
  "decision_type": "claim_approval",
  "decision_summary": "Approved claim #4821 based on medical necessity score 0.91",
  "risk_level": "medium",
  "compliance_tags": ["EU_AI_ACT_ART_13", "NAIC_AA"],
  "metadata": {
    "model": "claude-opus-4-6",
    "confidence": 0.91,
    "claim_id": "4821"
  }
}

evaluate_action

{
  "agent_id": "payment-agent",
  "action_type": "send_payment",
  "action_payload": {
    "amount": 125000,
    "currency": "USD",
    "recipient_type": "third_party"
  },
  "context": {
    "delegation_tier": 2,
    "jurisdiction": "US-CA"
  }
}

Returns: { "decision": "proceed" | "escalate" | "block", "reason": "...", "latency_ms": 42 }

get_compliance_matrix

{ "framework": "EU_AI_ACT" }

Returns an array of controls, each with status, evidence_count, last_evidence_date, and gap_description.

generate_evidence_package

{
  "framework": "EU_AI_ACT",
  "date_from": "2025-01-01",
  "date_to": "2025-03-31"
}