概要
What is Nest Llm Aigent?
Nest Llm Aigent is a forwarding/proxy solution that wraps multiple MCP (Model Context Protocol) servers into a unified HTTP service for large language models. Designed for developers integrating MCP capabilities into existing web services (e.g., NestJS), it acts as an AI agent that centralizes configuration and exposes standardized RESTful endpoints.
How to use Nest Llm Aigent?
Install MCP servers as private NPM packages and configure them in an mcp.config.json file. The running service exposes POST endpoints at paths like /api/mcp/tools, /api/mcp/tools/call, and /api/mcp/agent for HTTP-based access to tools, resources, and LLM invocation.
Key features of Nest Llm Aigent
- Acts as an AI agent aggregating multiple MCP servers
- Exposes standardized HTTP RESTful API endpoints
- Supports nearly all large language models
- MCP servers deployed as private NPM packages
- Centralized configuration via
mcp.config.json - Returns OpenAI-compatible function call definitions
Use cases of Nest Llm Aigent
- Integrate MCP tools into existing NestJS web services
- Centrally manage and version control multiple MCP servers
- Provide HTTP-based LLM access to non-MCP applications
- Unify tool, resource, and prompt retrieval in one service
FAQ from Nest Llm Aigent
What transport protocol does Nest Llm Aigent use for MCP servers?
Currently only stdio transport is implemented; SSE server support is planned for future versions.
How are MCP servers deployed and configured?
MCP servers are packaged as private NPM packages, installed via npm install, and configured in a mcp.config.json file defining server path, arguments, and name.
What endpoints does Nest Llm Aigent expose?
POST endpoints at /api/mcp/tools, /api/mcp/functools, /api/mcp/tools/call, /api/mcp/resources, /api/mcp/prompts, /api/mcp/all, and /api/mcp/agent.
Does Nest Llm Aigent support all large language models?
Yes, the project aims to support nearly all large model invocations through its HTTP API.
Where does tool execution happen?
Nest Llm Aigent is a forwarding layer; actual tool execution occurs in the connected MCP servers, which it bridges to HTTP requests.
