Composio powers 1000+ toolkits, tool search, context management, authentication, and a sandboxed workbench to help you build AI agents that turn intent into action.
Mirrored from https://github.com/Camo67/composio.git
Synced Nov 21, 04:45 UTC
| 1 | import { openai } from '@ai-sdk/openai'; |
| 2 | import { createMCPClient } from '@ai-sdk/mcp'; |
| 3 | import { Composio } from '@composio/core'; |
| 4 | import { VercelProvider } from '@composio/vercel'; |
| 5 | import { stepCountIs, streamText } from 'ai'; |
| 6 | |
| 7 | // 1. Initialize Composio. |
| 8 | const composio = new Composio({ |
| 9 | provider: new VercelProvider(), |
| 10 | }); |
| 11 | |
| 12 | // 2. Create an MCP session |
| 13 | console.log('🔄 Creating toolrouter session...'); |
| 14 | const session = await composio.create('default', { |
| 15 | toolkits: ['gmail'], |
| 16 | manageConnections: true, |
| 17 | mcp: true, |
| 18 | tools: { |
| 19 | gmail: { |
| 20 | enable: ['GMAIL_FETCH_EMAILS'], |
| 21 | }, |
| 22 | }, |
| 23 | }); |
| 24 | |
| 25 | const { mcp, sessionId } = session; |
| 26 | |
| 27 | console.log(JSON.stringify(mcp, null, 2)); |
| 28 | console.log(`✅ Toolrouter session created: ${sessionId}`); |
| 29 | |
| 30 | // 3. Create an MCP client |
| 31 | console.log(`🔄 Connecting to MCP Server: ${mcp.url}`); |
| 32 | const mcpClient = await createMCPClient({ |
| 33 | transport: { |
| 34 | type: 'http', |
| 35 | url: mcp.url, |
| 36 | headers: mcp.headers, |
| 37 | }, |
| 38 | }); |
| 39 | |
| 40 | // 4. Retrieve tools. |
| 41 | console.log(`🔄 Retrieving tools...`); |
| 42 | const tools = await mcpClient.tools(); |
| 43 | console.log(`✅ ${Object.values(tools).length} tools retrieved from MCP client`); |
| 44 | console.log(`🔎 Available tools: ${Object.keys(tools).join(', ')}`); |
| 45 | |
| 46 | // 5. Pass tools to Vercel-specific Agent. |
| 47 | console.log(`🔄 Executing agent...`); |
| 48 | const stream = streamText({ |
| 49 | model: openai('gpt-5.2'), |
| 50 | prompt: `Fetch my latest received email from Gmail and summarize it.`, |
| 51 | stopWhen: stepCountIs(10), |
| 52 | onStepFinish: step => { |
| 53 | if (step.toolCalls.length > 0) { |
| 54 | for (let i = 0; i < step.toolCalls.length; i++) { |
| 55 | const toolCall = step.toolCalls[i]; |
| 56 | console.log(`🔧 Executed ${toolCall.toolName}`); |
| 57 | const toolResult = step.toolResults?.[i]; |
| 58 | if (toolResult !== undefined) { |
| 59 | console.log(`✅ Result for ${toolCall.toolName}:`, toolResult); |
| 60 | } |
| 61 | } |
| 62 | } |
| 63 | }, |
| 64 | // output: Output.object({ |
| 65 | // schema: z.object({ |
| 66 | // subject: z.string(), |
| 67 | // bodySummary: z.string(), |
| 68 | // date: z.string(), |
| 69 | // sender: z.string(), |
| 70 | // }), |
| 71 | // }), |
| 72 | tools, |
| 73 | }); |
| 74 | |
| 75 | // 6. Execute the Vercel AI-specific Agent. |
| 76 | console.log(`🤖 Agent Response:`); |
| 77 | for await (const textPart of stream.textStream) { |
| 78 | process.stdout.write(textPart); |
| 79 | } |
| 80 | |
| 81 | process.stdout.write('\n'); |
| 82 | |
| 83 | // 7. Close Vercel AI's MCP client. |
| 84 | await mcpClient.close(); |