# oaid-mcp

MCP (Model Context Protocol) server that lets AI agents securely use their [Open Agent ID](https://openagentid.org) credential. The agent can sign requests, look up other agents, check credit scores, and exchange messages -- all without ever seeing the private key.

## Install

```bash
pip install oaid-mcp
```

## Configure

Set the path to your agent credential file:

```bash
export OAID_CREDENTIAL_FILE=/path/to/agent-credential.json
```

The credential file is generated by the `agent-id` CLI or SDK when you register an agent on-chain.

## Use with Claude Code

Add to your `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "oaid": {
      "command": "oaid-mcp",
      "env": {
        "OAID_CREDENTIAL_FILE": "/path/to/agent-credential.json"
      }
    }
  }
}
```

## Security

The private key is loaded by the MCP server process and **never** exposed to the AI agent. All cryptographic operations (signing, encryption) happen inside the server. The agent only receives DIDs, public info, and operation results.

## Available tools

| Tool | Description |
|------|-------------|
| `oaid_whoami` | Returns the agent's DID and public info (no private key) |
| `oaid_sign_request` | Signs an HTTP request and returns authentication headers |
| `oaid_check_credit` | Checks any agent's credit score via the registry |
| `oaid_lookup_agent` | Looks up agent info by DID |
| `oaid_send_message` | Sends a signed message to another agent |
| `oaid_get_messages` | Gets messages addressed to this agent |
| `oaid_send_encrypted_message` | Sends an end-to-end encrypted message to another agent |

## License

Apache-2.0