MCP Overview

Use MCP when the boundary is model-to-tool inside an application or local workstation. Use A2A when the boundary is agent-to-agent across organizations or services.

Choose The MCP Path

There are three supported ways to use JACS with MCP today:

Run jacs mcp when you want a ready-made MCP server with the broadest tool surface.
Wrap an existing MCP transport when you already have an MCP server or client and want signed JSON-RPC.
Register JACS as MCP tools when you want the model to call signing, verification, agreement, A2A, or trust operations directly.

Best Fit By Runtime

Runtime	Best starting point	What it gives you
Rust	`jacs-mcp`	Full MCP server with document, agreement, trust, A2A, and audit tools
Python	`jacs.mcp` or `jacs.adapters.mcp`	Local SSE transport security or FastMCP tool registration
Node.js	`@hai.ai/jacs/mcp`	Transport proxy or MCP tool registration for existing SDK-based servers

Important Constraints

Python MCP wrappers are local-only. JACSMCPClient, JACSMCPServer, and jacs_call() enforce loopback URLs.
Unsigned fallback is off by default. Both Python and Node fail closed unless you explicitly allow unsigned fallback.
Node has two factories. createJACSTransportProxy() takes a loaded JacsClient or JacsAgent; createJACSTransportProxyAsync() is the config-path variant.

1. Ready-Made Server: `jacs mcp`

Install the unified binary and start the MCP server:

cargo install jacs-cli
jacs mcp

The MCP server is built into the jacs binary (stdio transport only, no HTTP). It includes document signing, agreements, trust store operations, A2A tools, and security audit tools. See jacs-mcp/README.md in the repo for the full tool list and client configuration examples.

2. Transport Security Around Your Existing MCP Code

Python

Use jacs.mcp when you already have a FastMCP server or client and want transparent signing around the SSE transport:

from fastmcp import FastMCP
from jacs.mcp import JACSMCPServer

mcp = JACSMCPServer(FastMCP("Secure Server"), "./jacs.config.json")

For clients:

from jacs.mcp import JACSMCPClient

client = JACSMCPClient("http://localhost:8000/sse", "./jacs.config.json")

Helpful utilities in the same module:

create_jacs_mcp_server() for a one-line FastMCP server
jacs_middleware() for explicit Starlette middleware wiring
jacs_call() for one-off authenticated local calls

See Python MCP Integration for the detailed patterns.

Node.js

Use the transport proxy when you already have an MCP transport:

import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
import { JacsClient } from '@hai.ai/jacs/client';
import { createJACSTransportProxy } from '@hai.ai/jacs/mcp';

const client = await JacsClient.quickstart({
  name: 'mcp-agent',
  domain: 'mcp.local',
});

const transport = new StdioServerTransport();
const secureTransport = createJACSTransportProxy(transport, client, 'server');

If you only have a config path:

import { createJACSTransportProxyAsync } from '@hai.ai/jacs/mcp';

const secureTransport = await createJACSTransportProxyAsync(
  transport,
  './jacs.config.json',
  'server',
);

See Node.js MCP Integration for examples and tool registration.

3. Register JACS Operations As MCP Tools

This is different from transport security. Here the model gets explicit MCP tools such as jacs_sign_document, jacs_verify_document, agreement helpers, and trust helpers.

Python

from fastmcp import FastMCP
from jacs.client import JacsClient
from jacs.adapters.mcp import (
    register_jacs_tools,
    register_a2a_tools,
    register_trust_tools,
)

client = JacsClient.quickstart(name="mcp-agent", domain="mcp.local")
mcp = FastMCP("JACS Tools")
register_jacs_tools(mcp, client=client)
register_a2a_tools(mcp, client=client)
register_trust_tools(mcp, client=client)

Node.js

import { Server } from '@modelcontextprotocol/sdk/server/index.js';
import { JacsClient } from '@hai.ai/jacs/client';
import { registerJacsTools } from '@hai.ai/jacs/mcp';

const server = new Server(
  { name: 'jacs-tools', version: '1.0.0' },
  { capabilities: { tools: {} } },
);

const client = await JacsClient.quickstart({
  name: 'mcp-agent',
  domain: 'mcp.local',
});

registerJacsTools(server, client);

The Node tool set is intentionally smaller than the Rust MCP server. Use jacs mcp when you need the largest supported MCP surface.

Example Paths In This Repo

jacs-mcp/README.md
jacspy/examples/mcp/server.py
jacspy/examples/mcp/client.py
jacsnpm/examples/mcp.stdio.server.js
jacsnpm/examples/mcp.stdio.client.js

JACS Usage Documentation