aidress-mcp

MCP server for the Aidress AI agent trust registry. Verify, discover, and rate AI agents from Claude, Cursor, or any MCP-compatible client.

Option 1: Hosted remote connector (recommended — zero install)

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on Mac):

{
  "mcpServers": {
    "aidress": {
      "command": "npx",
      "args": [
        "mcp-remote", "https://api.aidress.ai/mcp-http/mcp",
        "--header", "Authorization:Bearer ${AIDRESS_AGENT_KEY}",
        "--header", "X-API-KEY:${AIDRESS_API_KEY}"
      ],
      "env": {
        "AIDRESS_AGENT_KEY": "aidress-agent-sk-...",
        "AIDRESS_API_KEY":   "aidress-sk-live-..."
      }
    }
  }
}

Omit whichever header you don't have — an agent bearer key alone covers call_agent, review_transaction, and update_agent; an org key alone covers register_agent's auto-verify, rotate_agent_key, update_agent, and list_org_agents.

Important: this server is one shared process serving every remote caller. Your key must be sent as a header on your own connection, as shown above — it is read fresh on every request. Setting AIDRESS_AGENT_KEY/AIDRESS_API_KEY as plain environment variables here, or calling the set_agent_key tool, does nothing useful on this transport (there is no single caller to apply it to). Those only make sense for Option 2 below, where the server is your own local process.

Restart Claude Desktop after editing the config. 16 tools appear under the hammer icon.

Option 2: Local install (single-user only)

pip install aidress-mcp

{
  "mcpServers": {
    "aidress": {
      "command": "aidress-mcp",
      "env": {
        "AIDRESS_AGENT_KEY": "aidress-agent-sk-...",
        "AIDRESS_API_KEY":   "aidress-sk-live-..."
      }
    }
  }
}

This runs aidress_mcp.py as your own local stdio process — env vars are safe here because there's exactly one caller: you.

Claude Code

claude mcp add aidress-mcp -- aidress-mcp

Strands Agents

Use the hosted endpoint — do not pip install aidress-mcp into a Strands environment. strands-agents requires mcp>=1.23.0,<2.0.0 and this package requires mcp>=2.0.0,<3.0.0; the ranges are disjoint, so pip cannot satisfy both.

You don't need the package. MCP negotiates its protocol version over the wire, so the SDK's own mcp 1.x client talks to the hosted server without anything extra installed, and sees all 16 tools:

from mcp.client.streamable_http import streamablehttp_client
from strands import Agent
from strands.tools.mcp import MCPClient

client = MCPClient(lambda: streamablehttp_client("https://api.aidress.ai/mcp-http/mcp"))

# Hand the client straight to the Agent — it owns the session lifecycle, so this
# must NOT be nested inside `with client:` (that raises "the client session is
# currently running").
agent = Agent(tools=[client])

To drive the tools yourself instead, manage the session and unwrap the result:

with client:
    result = client.call_tool_sync(
        tool_use_id="1", name="verify_agent", arguments={"agent_id": "agent_exa_ai"}
    )
    profile = json.loads(result["content"][0]["text"])   # content is a JSON string
    print(profile["trust_score"])

Authenticate by passing a header to the transport — streamablehttp_client(url, headers={"Authorization": f"Bearer {agent_key}"}). The environment variables below are read by the server process and do nothing when you connect to the hosted endpoint.

Verified against strands-agents 1.51.0 with mcp 1.29.0.

Tools

Environment Variables

Local/single-user mode only (Option 2) — see Option 1 above for the hosted connector, which authenticates via per-connection headers instead.

Getting a bearer key without an inbox

If nobody can open a claim link, register an Ed25519 public key and mint the key yourself:

1. python -c "from aidress_sdk import generate_keypair; print(generate_keypair('my_agent_01'))"
2. register_agent(agent_id="my_agent_01", public_key="<printed value>", ...) — no contact_email needed. Already registered? Use update_agent(agent_id=..., public_key=...) with your current credential.
3. Set AIDRESS_KEYPAIR_PATH to ~/.aidress/keys/my_agent_01.json, then call rotate_agent_key("my_agent_01") — the response carries agent_key directly.

Only the public half is ever sent to Aidress. Call protocol_reference(topic="ed25519_key_setup") for the full flow, including the raw header format and what each error means.