MCP Flow – Unified Chat Workflow Engine

MCP Flow is a lightweight, Python‑based engine that standardizes the interaction between AI assistants and external services through the Model Context Protocol. It solves a common pain point for developers: disparate tools, inconsistent data pipelines, and ad‑hoc request handling that make it hard to build reliable conversational applications. By exposing a single, well‑defined MCP endpoint, MCP Flow lets AI assistants such as Claude or OpenAI’s agents send structured requests and receive deterministic responses, regardless of the underlying logic.

What It Does

At its core, MCP Flow orchestrates a sequence of conversational steps—formatting user input, generating answers, and delegating to custom ADK agents. It implements these steps as workflows, each represented by a JSON schema that describes the expected input, output, and intermediate transformations. The engine runs on FastAPI, providing a RESTful interface that can be called by any MCP‑compatible client. A command‑line tool allows developers to fire test requests locally, inspect logs, and iterate quickly.

Key Features

• Multi‑workflow support – Formatters, answer generators, and ADK agents can be combined in any order.
• ADK integration – Leverages Google’s Agent Development Kit to create custom agents and plug‑in tools that can be invoked directly from the workflow.
• Extensibility – New workflows or plug‑ins can be added without touching the core engine, making it future‑proof.
• CLI testing – A simple command line interface speeds up debugging and prototyping.

Use Cases

• Customer support bots that need to retrieve data from internal APIs before answering.
• Knowledge‑base assistants that format queries, call an LLM for a draft response, and then pass it to an ADK agent that validates facts.
• Data‑centric applications where a user’s request triggers multiple downstream services (e.g., fetching weather, stock prices, and then summarizing the result).

Integration with AI Workflows

MCP Flow plugs into existing MCP pipelines by acting as a tool endpoint. An AI assistant sends a structured request that specifies the workflow name and payload; MCP Flow executes the defined steps, returning a clean JSON response. Because it follows the MCP schema, the assistant can automatically chain calls—e.g., after formatting, pass the result to an ADK agent, then back to the assistant for final delivery. This seamless handoff reduces boilerplate code and keeps the conversational logic declarative.

Unique Advantages

• Unified protocol – Eliminates the need for custom adapters per tool.
• FastAPI + ADK combo – Combines rapid API development with powerful agent capabilities in a single stack.
• Developer‑friendly tooling – The CLI and clear workflow definitions make it easy to prototype, test, and extend without deep infrastructure knowledge.

MCP Flow thus empowers developers to build robust, maintainable conversational systems that can scale from simple Q&A bots to complex, multi‑step reasoning pipelines—all while staying compliant with the Model Context Protocol.