Model Context Protocol moderate risk

Acp

The ACP to MCP adapter bridges ACP agents to MCP applications, enabling MCP tools to discover and run ACP agents as resources.

Connections & Capabilities

Connects To

GitHubDiscordDocker

Capabilities

readexec

Quickstart

Install

uvx acp-mcp http://localhost:8000

Config

{
  "mcpServers": {
    "acp-local": {
      "command": "uvx",
      "args": ["acp-mcp", "http://localhost:8000"]
    }
  }
}

Exposed MCP Tools (1)

moderate

run_agent

Invokes a specified ACP agent with a given input.

The agent's actions depend on its implementation, which could include non-destructive writes.

Safety Assessment

The adapter allows MCP applications to run ACP agents, which can be risky depending on the agent's capabilities. It is relatively safe when using well-understood, read-only ACP agents. It becomes risky when agents have write or execute permissions, especially if the agent's behavior is not well-defined or trusted.

ACP agents are exposed as MCP resources, limiting direct access.
Requires explicit user acceptance within the MCP client to enable integration.
Adapter operates within the constraints of the MCP client application.
No direct internet exposure of ACP agents via the adapter.

ACP agents are executed as tools, potentially performing actions defined in their implementation.
Basic content translation might introduce vulnerabilities if not handled carefully.
Lack of shared memory could lead to inconsistent state if agents manage external resources.
Potential for information leakage if ACP agent responses contain sensitive data.