Viterbit MCP Server

## ChatGPT Custom GPT Setup for Viterbit MCP ### Issue: ChatGPT Schema Validation ChatGPT validates requests **client-side** against the OpenAPI schema before sending them to your server. If the schema is too strict, it will reject requests that don't match perfectly. ### Solution: Use the Simplified OpenAPI Schema Copy this **exact** schema into your ChatGPT Custom GPT Action: ```yaml openapi: 3.1.0 info: title: Viterbit MCP Server description: Access to Viterbit recruitment API version: 2.0.0 servers: - url: https://mcp-viterbit.onrender.com paths: /tools/call: post: operationId: callTool summary: Execute a Viterbit tool security: - ApiKeyAuth: [] requestBody: required: true content: application/json: schema: type: object responses: '200': description: Success content: application/json: schema: type: object components: securitySchemes: ApiKeyAuth: type: apiKey in: header name: X-API-Key ``` **Key Points:** - `schema: type: object` with NO properties defined = accepts any JSON - This lets ChatGPT send requests in any format - Your server handles the parsing flexibly ### Setup Steps 1. **Go to ChatGPT** → My GPTs → Your Viterbit GPT 2. **Configure** → **Actions** → **Edit Action** 3. **Replace the entire schema** with the simple one above 4. **Update your server URL** in the schema: - Replace `https://mcp-viterbit.onrender.com` with your actual Render URL 5. **Authentication:** - Type: `API Key` - Header name: `X-API-Key` - Value: One of your `MCP_API_KEYS` 6. **Save** ### Custom GPT Instructions Add these instructions to help ChatGPT format requests correctly: ``` You are a Viterbit recruitment assistant with access to the Viterbit API. When calling tools: - Use the callTool operation - Send requests in this format: {"name": "tool_name", "arguments": {"param": "value"}} - Available tools: search_candidate, get_candidate_details, update_candidate_discord_id, etc. - Always check the result for success/error before responding to the user Example tool call for searching a candidate: { "name": "search_candidate", "arguments": { "email": "candidate@example.com" } } If a tool call fails, explain the error to the user clearly. ``` ### Testing Once configured, test with: ``` Search for candidate with email: fernandezporrasluisalberto@gmail.com ``` ChatGPT should call: ```json { "name": "search_candidate", "arguments": { "email": "fernandezporrasluisalberto@gmail.com" } } ``` ### Troubleshooting **Error: "Error al hablar con connector"** - This means ChatGPT rejected the request CLIENT-SIDE - The request never reached your server - Fix: Use the simplified schema above **Error: "Missing required parameter"** - The request reached your server but parameters are missing - Check ChatGPT's request format in the action logs - The server expects either: - `{"name": "tool", "arguments": {"param": "val"}}` - OR `{"name": "tool", "param": "val"}` **Error: "Invalid API key"** - Check your X-API-Key matches one in `MCP_API_KEYS` - Verify authentication is configured correctly in ChatGPT ### Alternative: List All Tools First Add this tool to your OpenAPI schema to let ChatGPT discover available tools: ```yaml /tools: get: operationId: listTools summary: Get all available Viterbit tools security: - ApiKeyAuth: [] responses: '200': description: List of tools with schemas content: application/json: schema: type: array ``` Then instruct ChatGPT: ``` At the start of each conversation, call listTools to see what's available. ``` ### Server Logs To debug, check your Render logs: - Go to Render Dashboard - Click your service - Click "Logs" - Look for: - `Raw request body: ...` (shows exactly what was sent) - `Tool called: ... with arguments: ...` - Any error messages The server logs EVERYTHING for debugging.