remote-mcp-agent

# Inline Agents Worker Configuration This repository contains a simple Lambda function setup using LangChain to dynamically create an AI agent that consumes remote MCP servers. ## Remote MCP Server The code for the MCP server is contained in the [mcp folder](/mcp). This server uses Cloudflare for hosting. **You must have a Cloudflare account to deploy it**. Once your account is created and logged in, you can run the following commands to deploy the worker into Cloudflare: ```bash cd mcp npm run deploy ``` ## Lambda functions This solution also deploys four Lambda functions into AWS for agent support using SAM. To deploy the resources into your AWS account you can use the following commands: ```bash sam build sam deploy --guided ``` This will walk you through the deployment. Once deployed, you will have access to a public API that allows you to configure MCP servers for the agent to consume. ### AI Agent The [AI agent](/functions/agent.mjs) uses LangChainjs to create an inline agent and give it the configured MCP servers. It uses OpenAI as the LLM to perform the tasks. ## Endpoints The base url for your deployment is an output of the stack. You can prepend that to the following endpoints: `POST /agent` - executes the agent and provides it a prompt passed in from the `message` body variable `GET /servers` - returns an html page allowing you to configure MCP servers for your agent to consume (only SSE/remote servers are supported) `POST /servers` - adds or updates a remote MCP server for your agent. This is called via the html page `DELETE /servers/{name}` - removes a remote MCP server from your agent configuration