MCP Server Configuration

Each MCP server is defined as a named section under [mcp] in config.toml.

Config Format

[mcp.<server-name>]
command = "<executable>"
args = ["arg1", "arg2", ...]

Discovery Process

1. For each [mcp.*] section, spawn the process with the configured command and args
2. Initialize the JSON-RPC connection (protocol handshake)
3. Call tools/list to enumerate available tools
4. Wrap each tool's name, description, and input schema as a native Tool
5. Log any errors per-server without failing globally

Examples

Filesystem Access

[mcp.filesystem]
command = "npx"
args = ["-y", "@anthropic/mcp-filesystem", "/home/user/projects"]

Web Search

[mcp.brave-search]
command = "npx"
args = ["-y", "@anthropic/mcp-brave-search"]

Custom Python Server

[mcp.my-server]
command = "python3"
args = ["/path/to/my_mcp_server.py"]

Threat Detection

aidaemon performs audit-only threat detection on MCP tool calls. Suspicious patterns are logged but do not block execution.

Suspicious Argument Patterns

• File access: /etc/passwd, /etc/shadow, .ssh/, .env
• Config/secrets: config.toml, aidaemon.db, api_key, bot_token, encryption_key
• Network: curl, wget, nc, base64
• Code execution: eval(, exec(, | sh, | bash
• Destructive: ; rm , chmod 777

Suspicious Output Patterns

• Potential API keys: sk-, ghp_ prefixes
• Private keys: -----BEGIN, PRIVATE KEY
• Sensitive terms: password, bot_token