agentmesh_mcp_trust

MCP Trust Server for AgentMesh

MCP server that exposes AgentMesh trust management as tools for AI agents via the Model Context Protocol. Compatible with Claude, GPT, and any MCP-capable client.

What it does

Provides six trust-management tools over MCP:

Trust is scored across multiple dimensions (0–1000 each).

Getting Started

Prerequisites: Python 3.11 or newer, pip, and an MCP-capable client such as Claude Desktop.

python -m venv .venv
source .venv/bin/activate
pip install mcp-trust-server
AGENTMESH_AGENT_NAME=my-trust-server mcp-trust-server

The server starts with in-memory storage by default. Set AGENTMESH_MIN_TRUST_SCORE or AGENTMESH_STORAGE_BACKEND before launch to customize the local policy.

Installation

pip install mcp-trust-server

Or install from the repository:

cd agent-governance-python/agent-mesh/packages/mcp-trust-server
pip install -e ".[dev]"

Configuration

Usage

Run directly

python -m mcp_trust_server

Or via the console script:

mcp-trust-server

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "agentmesh-trust": {
      "command": "python",
      "args": ["-m", "mcp_trust_server"],
      "env": {
        "AGENTMESH_AGENT_NAME": "my-trust-server",
        "AGENTMESH_MIN_TRUST_SCORE": "500"
      }
    }
  }
}

Using with pip install

{
  "mcpServers": {
    "agentmesh-trust": {
      "command": "mcp-trust-server"
    }
  }
}

Tool Details

check_trust(agent_did: str) -> dict

Quick trust check. Returns whether the agent meets the minimum trust threshold, overall score, and trust level.

get_trust_score(agent_did: str) -> dict

Full trust breakdown — overall score, trust level, interaction count, and last-updated timestamp.

establish_handshake(peer_did: str, capabilities: list[str]) -> dict

Creates a challenge for a trust handshake. Returns handshake ID, signature, and status.

verify_delegation(agent_did: str, delegator_did: str, capability: str) -> dict

Validates that a delegation from delegator_did to agent_did for the given capability is trustworthy.

record_interaction(peer_did: str, outcome: str, details: str) -> dict

Records an interaction with a peer and adjusts trust scores. Valid outcomes: success (+10), failure (−20), timeout (−10), partial (+5).

get_identity() -> dict

Returns this server instance's DID, name, Ed25519 public key, and granted capabilities.

Example Interaction

User: Check if agent did:mesh:abc123 is trusted

Claude calls check_trust(agent_did="did:mesh:abc123")
→ {
    "agent_did": "did:mesh:abc123",
    "trusted": true,
    "overall_score": 500,
    "trust_level": "standard",
    "dimensions": {
      "competence": 500,
      "integrity": 500,
      "availability": 500,
      "predictability": 500,
      "transparency": 500
    },
    "min_trust_threshold": 500
  }

Development

cd agent-governance-python/agent-mesh/packages/mcp-trust-server
pip install -e ".[dev]"
pytest tests/ -v

License

Apache-2.0