Extensibility via Model Context Protocol (MCP)
Connecting your agents to the world, without the wrapper hell.
The Integration Gap
In the Hermes ecosystem, adding a new tool or data source usually means writing a custom Python wrapper, managing local dependencies, and hoping the agent's tool-calling logic doesn't hallucinate the arguments. With the broken skill downloads (#5053), even the official way to extend Hermes is currently compromised.
Gobii v2.19.0 introduces native support for Remote MCP Server Endpoints. This allows you to connect Gobii agents to any tool, database, or API that speaks the MCP standard — hosted anywhere.
What is MCP?
The Model Context Protocol (MCP) is an open standard that enables AI models to interact with external data and tools in a structured, predictable way. Instead of writing "glue code" for every integration, you simply point your agent to an MCP server.
Hermes Local Wrappers vs. Gobii Remote MCP
| Feature | Hermes (Local Wrappers) | Gobii (Remote MCP) |
|---|---|---|
| Setup Complexity | High (Custom Python) | Low (URL Endpoint) |
| Dependency Management | Manual (pip/conda) | Server-side (Isolated) |
| Extensibility | Limited by local env | Unlimited (Remote) |
| Maintenance | You own the code | Standardized protocol |
| Security | Local execution risk | Sandboxed/Remote isolation |
The Power of Remote Endpoints
By supporting remote MCP endpoints, Gobii allows for a truly modular agent architecture. You can host your sensitive data tools on a private server and expose them to your Gobii agents via a secure MCP bridge. No need to move data into the agent's local environment, and no need to rebuild the agent every time a tool changes.