TrustMesh MCP Server

Model Context Protocol server for TrustMesh — AI agent trust & reputation scoring.

Quick Start

npx trustmesh-mcp

Or install globally:

npm install -g trustmesh-mcp
trustmesh-mcp

Configuration

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "trustmesh": {
      "command": "npx",
      "args": ["trustmesh-mcp"]
    }
  }
}

Cursor / Windsurf

Add to your MCP settings:

{
  "mcpServers": {
    "trustmesh": {
      "command": "npx",
      "args": ["trustmesh-mcp"],
      "env": {
        "TRUSTMESH_NODE_ID": "my-agent-1"
      }
    }
  }
}

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | TRUSTMESH_NODE_ID | mcp-<random> | Your agent's identity in the mesh | | TRUSTMESH_REMOTE_URL | local only | URL of a remote TrustMesh API server | | TRUSTMESH_API_KEY | none | API key for remote server auth |

Tools

rate

Rate an interaction with another AI agent.

Parameters:

• target (string, required) — Agent ID to rate
• score (number 0-1, required) — Trust score
• claim (string, optional) — Description of the interaction
• category (enum, optional) — reliability, quality, timeliness, fairness, security

score

Get the trust score for an agent.

Parameters:

• agentId (string, required) — Agent ID to look up

batch_score

Get trust scores for multiple agents at once (max 100).

Parameters:

• agentIds (string[], required) — List of agent IDs

graph

Get the trust graph around an agent.

Parameters:

• agentId (string, required) — Agent ID to inspect
• depth (number 1-3, default 1) — Graph traversal depth

dispute

Create a dispute against a rating.

Parameters:

• ratingId (string, required) — The rating to dispute
• reason (string, required) — Why the rating is disputed

check_before_trust

Quick trust check — returns proceed, caution, or avoid.

Parameters:

• agentId (string, required) — Agent to check
• threshold (number 0-1, default 0.6) — Minimum trust score

my_score

Get your own trust score as this MCP node.

Resources

• trustmesh://config — Current node configuration

How It Works

TrustMesh uses peer-to-peer trust scoring: agents rate each other after interactions, and scores are computed using a weighted algorithm that factors in rater credibility, recency, and category.

When running locally (no TRUSTMESH_REMOTE_URL), all data is stored in memory and doesn't persist between sessions. For production use, point to a remote TrustMesh API server.

Links

• npm: https://www.npmjs.com/package/trustmesh-mcp
• GitHub: https://github.com/geerdwedda-create/trustmesh
• Core SDK: https://www.npmjs.com/package/trustmesh-ai
• Docs: https://geerdwedda-create.github.io/trustmesh/

License

MIT