taxi-callcenter-agents

Taxi Callcenter Agents is a demo MCP server that coordinates a Google ADK-based call center agent for taxi ordering. It exposes a server component you can run locally or in the cloud and a client toolchain to interact with the MCP. The repository illustrates building and running the agent, and provides an example MCP configuration to connect a local MCP server (gcloud) to the agent, enabling standard I/O transport for communication. You can also explore the MCP introspector for easy local development, which lets you run the inspector as a stdio binary or as an SSE server to inspect and test MCP interactions. The end-to-end flow typically involves launching the MCP server, starting the agent UI or API, and connecting via the MCP protocol to exchange messages and orchestrate taxi ordering via Google ADK.

Prerequisites:

• Go (for building or running via Go sources)
• goreleaser (for building release snapshots)
• uv (or an alternative tool) as shown in the project setup steps

1. Install prerequisites

• Install Go from https://golang.org/dl/
• Install goreleaser following https://goreleaser.com/install/
• Install uv (as shown in the project setup) or ensure your environment provides uv

2. Build or prepare the MCP server (taxi-callcenter-agents)

• If you want a release snapshot build (as shown in the README): goreleaser build --snapshot --clean
• To create a release: goreleaser release --skip-publish --snapshot --clean

3. Run the local MCP server example

• For the local gcloud-like setup (using stdio transport): Ensure you have the built MCP server binary at the path you specify in the mcp_config (see the example in the README). Use the provided config snippet to connect via stdio transport.

4. Quick start with MCP introspector (optional)

• As stdio binary: npx @modelcontextprotocol/inspector go run main.go
• As SSE server: go run main.go --transport sse npx @modelcontextprotocol/inspector npx mcp-remote@next http://localhost:8001/sse npx @modelcontextprotocol/inspector

5. Cloud Run deployment (optional)

• Follow the Cloud Run deployment steps in the README to deploy the MCP server and the simple taxi agent, including enabling required Google APIs and setting environment variables like BASE_URL, MCP_SERVER_URL, GOOGLE_API_KEY, and GOOGLE_GENAI_USE_VERTEXAI as appropriate.

Notes and tips:

• The example MCP configuration uses a stdio transport, which is common for local development and debugging. Replace the command path with the actual path to your built MCP server binary when deploying.
• Environment variables (env) are important for real deployments. For example, enable BASE_URL for the server URL, and provide API keys or feature flags (e.g., GOOGLE_API_KEY, GOOGLE_GENAI_USE_VERTEXAI) as needed by the agent.
• When deploying to Cloud Run or other cloud environments, ensure the MCP server and agent have network access to each other and that the port mappings (e.g., 8001 for the MCP server and 8000 for the UI) match your deployment configuration.
• The README shows multiple ways to run and test: local standard I/O transport, inspection tools, and cloud deployments. Choose the method that matches your development or production workflow.
• If you switch from stdio to SSE or other transports, update the mcpServers configuration accordingly (transport type may affect how you pipe data between the MCP server and agent).
• For troubleshooting, consult the container or process logs in your deployment environment and verify that the MCP server URL (MCP_SERVER_URL) and UI keys are correctly set.