Skip to main content
The Agent Connect Framework (ACF) is designed to facilitate collaboration between agents, regardless of the agent framework or deployment platform. This framework is part of a larger ecosystem of standards and frameworks that address similar concerns. The ultimate goal of ACF and the IBM watsonx platform is to be inclusive of open standards, open source, and open protocols that have proven value to the developer community and AI-powered ecosystems.

​
Model Context Protocol (MCP)

MCP is an open protocol that standardizes how applications provide context to large language models (LLMs). It provides a standardized way for AI models to connect to different data sources and tools, similar to how a USB-C port connects devices to peripherals. ACF does not define how agents connect to data sources or tools, thus being compatible with MCP. MCP uses a style of JSON-RPC for communication, which is optimized for local services and HTTP with support for server-sent events (SSE), enabling streaming of tool calls and responses. This communication makes MCP highly compatible with ACF, which assumes steps or actions that agents take happen asynchronously, and event messages relay responses and status back to the calling application or agent. For more information on MCP, see the following resources:
Find the most recent MCP specification here
Find the main MCP documentation here.

​
Agent Protocol

The Agent Protocol, defined by LangChain.ai, is a framework-agnostic API for serving LLM-based agents in production environments. It centers around three key concepts:
  • Runs: APIs for executing an agent.
  • Threads: APIs to organize multi-turn executions of agents.
  • Store: APIs to work with long-term memory.
This protocol is similar to the Agent API set defined by ACF. Collaboration between the watsonx Orchestrate and the LangChain teams is in progress to ensure cross-compatibility between ACF and Agent Protocol, with the ultimate goal of seamless bi-directional interoperability. Future versions of ACF may adopt an Agent Protocol flavor of the core Agent APIs to enable this compatibility. For more information on Agent Protocol, see the following resources:
Find the full OpenAPI specification here.
Find the JSON specification here.

​
AGNTCY

AGNTCY is an open-source collective building the infrastructure for the Internet of Agents: an open, interoperable internet for agent-to-agent collaboration. The Agent Connect Protocol within AGNTCY is heavily inspired by the Agent Protocol by LangChain. For more information on AGNTCY, see the following resources:
Find the AGNTCY website here.
Find the protocol specification here.

​
Google Agent2Agent Protocol (A2A)

Google has announced an open protocol called Agent2Agent (A2A), with support and contributions from over 50 technology partners. A2A enables AI agents to communicate with each other, securely exchange information, and coordinate actions on top of various enterprise platforms or applications. Similar to ACF, A2A’s goal is to enable AI agents to work across entire enterprise application estates. A2A addresses the following capabilities:
  • Capability discovery: Agents can advertise their capabilities using an “Agent Card” in JSON format, allowing the client agent to identify the best agent to perform a task and leverage A2A to communicate with the remote agent.
  • Task management: The communication between a client and remote agent is oriented towards task completion, in which agents work to fulfill end-user requests. The “task” object is defined by the protocol and has a lifecycle. It can be completed immediately or, for long-running tasks, each of the agents can communicate to stay in sync with each other on the latest status of completing a task. The output of a task is known as an “artifact.”
  • Collaboration: Agents can send each other messages to communicate context, replies, artifacts, or user instructions.
  • User experience negotiation: Each message includes “parts,” which is a fully formed piece of content, like a generated image. Each part has a specified content type, allowing client and remote agents to negotiate the correct format needed and explicitly include negotiations of the user’s UI capabilities like iframes, video, web forms, and more.
Google has published a draft spec here.