ACP to MCP Adapter

The ACP to MCP Adapter serves as a lightweight bridge that lets AI developers combine the strengths of two complementary ecosystems: the Agent Communication Protocol (ACP) and the Model Context Protocol (MCP). ACP is designed for agent‑to‑agent coordination, enabling autonomous agents to share data, trigger actions, and orchestrate complex workflows. MCP, on the other hand, is a protocol that allows AI models to discover and call external tools or resources in a structured way. By exposing ACP agents as MCP resources, the adapter eliminates the need to re‑implement agent logic for each new model or toolchain, allowing existing ACP agents to be leveraged directly within MCP‑enabled applications such as Claude Desktop.

At its core, the adapter translates ACP agent endpoints into MCP tools and resources. When an MCP client queries the server, it receives a list of available agents, each represented as a callable tool. Invoking one of these tools runs the corresponding ACP agent and streams the result back to the model, effectively turning an autonomous agent into a first‑class tool for the assistant. This seamless integration means developers can add sophisticated, stateful agent behaviors to conversational AI without modifying the underlying model or writing custom connectors.

Key capabilities include:

• Discoverability – ACP agents automatically appear in the MCP resource list, making them visible to any MCP client.
• Tool Exposure – Each agent run is exposed as a tool, allowing the model to call it with arbitrary input and receive structured output.
• Minimal Configuration – The adapter requires only a single command to start and can be launched locally or in Docker, making it easy to embed into existing development workflows.

Typical use cases span from simple echo or calculator agents to complex multi‑step reasoning pipelines where an agent orchestrates calls to external APIs, databases, or other services. For example, a customer‑support assistant could invoke an ACP agent that queries a CRM system, processes the data, and returns a concise summary for the model to present. Because the adapter preserves agent state across calls, developers can build long‑running workflows that maintain context without reinitializing the model.

The adapter’s design prioritizes interoperability and developer convenience. It removes the friction of porting agent logic into new toolchains, enabling rapid experimentation and deployment across diverse AI platforms. While it currently lacks streaming updates and deep data‑structure support, its lightweight nature and straightforward integration make it a powerful addition to any MCP‑based AI workflow.