MCP Integration

Zene is designed to be highly extensible through the Model Context Protocol (MCP) and modular tool definitions.

1. Model Context Protocol (MCP)

Zene adopts MCP as the standard for interacting with external resources and tools.

• Ecosystem Compatibility: Use any existing MCP server (Git, Postgres, Google Search) without custom glue code.
• Standardization: MCP defines clear schemas for Tools, Resources, and Prompts.
• Isolation: MCP servers run as independent processes (Stdio/HTTP), keeping Zene's core stable.

MCP Architecture

Zene acts as an MCP Client connecting to multiple MCP Servers (Sidecars).

graph LR
    Zene[Zene Core] -- Stdio --> S1[Git MCP]
    Zene -- Stdio --> S2[Postgres MCP]
    S1 --> Git[Git Binary]
    S2 --> DB[(Database)]

2. Extension Points

Beyond MCP, Zene can be extended internally:

• Custom Tools: Add new capabilities by defining tools in src/engine/tools.rs and registering them in the ToolManager.
• New Agent Types: Modify the Orchestrator or add specialized logic for different reasoning patterns.
• Custom Persistence: Implement the SessionStore trait to support new storage backends (e.g., S3, Redis).

3. Adding New Functionality

1. Identify the Point: Decide if it's a stand-alone MCP server or a core Zene tool.
2. Implement: Adhere to the ZeneError system and async patterns.
3. Configure: Add MCP servers to zene_config.toml:

   [mcp_servers]
   git = { command = "uvx", args = ["mcp-server-git"] }

4. Verify: Use the integration test suite to ensure the new capability is reachable by the Agent.