joshuamanivinod/twitter-agent-using-mcp

Created an AI agent to do twitter post on my behalf using mcp server

client

1. import readline/promises
2. use readline to create interface for i/o
3. create chatloop to ask question(promise)
4. push user,part:(text,type) question to history
5. create response using await
6. using @modelcontextprotocol/sdk import McpServer,SSEServerTransport
7. create mcpClient
8. declare tools=[] globally
9. connect mcpClient to backend("/sse") using SSEClientTransport
10. like this we get access to all the resourses prompts and tools of backend
11. map each tool with the help of console.log("Available tools: ", tools);
12. list the tools in chatloop
13. to tell AI hamare paas itne tools hai, aur tum uska use kar sakte ho, we add config while taking reponse using generateContent

    config:{
      tools:[
        {
          functionDeclarations: tools
        }
      ]
    }

14. shift calling chatloop() outside to inside mcpClient
15. whenever ai does a function call we get to see when console.log(response.candidates[0].content) or console.log(response.candidates[0].content.parts[0])
16. we need to allow ai to do the function call, how? by doing const functionCall=response.candidates[0].content.parts[0].functionCall and if a functionCall exists then call chatloop(functionCall) and in chatloop(toolCall) and then do mcpClient.callTool
17. push both Calling tool and toolResult to chatHistory
18. run node index.js

server

1. boiler plate (run npx nodemon index.js)

import express from "express";
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { SSEServerTransport } from "@modelcontextprotocol/sdk/server/sse.js";

const server = new McpServer({
    name: "example-server",
    version: "1.0.0"
});

const app = express();

const transports = {};

app.get("/sse", async (req, res) => {
    const transport = new SSEServerTransport('/messages', res);
    transports[ transport.sessionId ] = transport;
    res.on("close", () => {
        delete transports[ transport.sessionId ];
    });
    await server.connect(transport);
});

app.post("/messages", async (req, res) => {
    const sessionId = req.query.sessionId;
    const transport = transports[ sessionId ];
    if (transport) {
        await transport.handlePostMessage(req, res);
    } else {
        res.status(400).send('No transport found for sessionId');
    }
});

app.listen(3001, () => {
    console.log("Server is running on http://localhost:3001");
});

2. create McpServer tools
3. in server tool we pass func name,discription, schema, function
4. use zod to specify the schema
5. note: all functions inside zod are async
6. function return is a bit fancy
7. npm i twitter-api-v2
8. create file mcp.tool.js
9. create twitter account and give read and write access
10. get TWITTER_APP_KEY,TWITTER_APP_SECRET,TWITTER_ACCESS_TOKEN,TWITTER_ACCESS_SECRET and store them in .env
11. status contains what you are going to post which we get from client
12. run server using npx nodemon index.js