PQC MCP Transport

Post-quantum secured transport layer for the Model Context Protocol (MCP). Every JSON-RPC message is signed with ML-DSA (FIPS 204) digital signatures, providing cryptographic authentication, integrity verification, and replay protection that is resistant to both classical and quantum attacks.

Installation

pip install pqc-mcp-transport

For development:

pip install pqc-mcp-transport[dev]

Quick Start

Server

import asyncio
from quantumshield.identity.agent import AgentIdentity
from pqc_mcp_transport import PQCMCPServer

server_identity = AgentIdentity.create("my-server")
server = PQCMCPServer(identity=server_identity)

@server.tool("greet", description="Greet someone")
async def greet(name: str) -> str:
    return f"Hello, {name}!"

asyncio.run(server.run(port=8080))

Client

import asyncio
from quantumshield.identity.agent import AgentIdentity
from pqc_mcp_transport import PQCMCPClient

async def main():
    agent = AgentIdentity.create("my-client")
    client = PQCMCPClient(identity=agent, server_url="http://localhost:8080")

    session = await client.connect()      # PQC handshake
    result = await client.call_tool("greet", {"name": "World"})
    print(result)                          # Verified response
    await client.close()

asyncio.run(main())

Architecture

 Client                                          Server
 ------                                          ------
   |                                               |
   |  1. HandshakeRequest (signed with ML-DSA)     |
   |---------------------------------------------->|
   |                                               | verify client sig
   |  2. HandshakeResponse (signed with ML-DSA)    |
   |<----------------------------------------------|
   | verify server sig                             |
   |                                               |
   |  === Session Established (mutual auth) ===    |
   |                                               |
   |  3. JSON-RPC Request + _pqc envelope          |
   |---------------------------------------------->|
   |                                  verify sig,  | check nonce,
   |                                  execute tool | sign response
   |  4. JSON-RPC Response + _pqc envelope         |
   |<----------------------------------------------|
   | verify response sig                           |

Protocol Specification

Handshake (Mutual Authentication)

1. Client generates a nonce, signs {did}:{nonce}:{timestamp} with its ML-DSA private key, and sends a HandshakeRequest.
2. Server verifies the client's signature, generates its own nonce and a session ID, signs {did}:{client_nonce}:{server_nonce}:{session_id}, and returns a HandshakeResponse.
3. Client verifies the server's signature and the echoed nonce. A PQCSession is created on both sides.

Message Format

Every MCP JSON-RPC message carries a _pqc envelope:

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "id": "abc123",
  "params": { "name": "greet", "arguments": { "name": "World" } },
  "_pqc": {
    "signer_did": "did:pqaid:abcdef...",
    "algorithm": "ML-DSA-65",
    "timestamp": "2025-01-15T10:30:00+00:00",
    "nonce": "a1b2c3d4e5f6...",
    "signature": "3045022100...",
    "public_key": "302a300506..."
  }
}

The _pqc field is stripped before signing (canonical form) and before passing the message to standard MCP handlers.

Signing Process

1. Remove _pqc from the message.
2. Serialize with json.dumps(msg, sort_keys=True, separators=(',', ':')).
3. Compute SHA3-256 hash of the canonical bytes.
4. Sign the hash with ML-DSA.
5. Attach the _pqc envelope with signature, public key, nonce, and timestamp.

Security Properties

Property	Mechanism
Message Authentication	Every message is ML-DSA signed
Mutual Authentication	Both client and server verify each other during handshake
Integrity	Canonical JSON + SHA3-256 hash prevents tampering
Replay Protection	Per-session nonce tracking rejects duplicates
Session Expiry	Sessions have a configurable TTL (default: 1 hour)
Quantum Resistance	ML-DSA (FIPS 204) is resistant to Shor's algorithm
Audit Trail	Every operation is logged with signature metadata

API Reference

MessageSigner

Method	Description
canonicalize(message)	Deterministic JSON serialization (static)
sign_message(message)	Add _pqc envelope with ML-DSA signature
verify_message(message)	Verify _pqc envelope, returns VerificationResult (static)
strip_pqc(message)	Remove _pqc for standard MCP processing (static)

PQCHandshake

Method	Description
initiate(identity)	Create a signed handshake request
respond(request, server_identity)	Verify client, create signed response
complete(response, client_identity, nonce)	Verify server, create session

PQCSession

Method	Description
is_valid()	Check if session has not expired
check_nonce(nonce)	Register nonce, raise ReplayAttackError on reuse
log_operation(...)	Record operation in audit trail
get_audit_log()	Return list of AuditEntry records

PQCMCPClient

Method	Description
connect()	Perform PQC handshake, return PQCSession
call_tool(name, arguments)	Send signed tool call, verify response
list_tools()	List available tools (signed request)
close()	Close connection

PQCMCPServer

Method	Description
tool(name, description)	Decorator to register a tool handler
handle_request(raw_message)	Process incoming request with PQC verification
handle_handshake(request)	Handle handshake initiation
get_tool_list()	Return registered tools
run(host, port)	Start HTTP server

PQCMiddleware

ASGI middleware for adding PQC to existing frameworks (Starlette, FastAPI):

from pqc_mcp_transport.middleware import PQCMiddleware
app = PQCMiddleware(app, server_identity=identity)

Exceptions

Exception	When
PQCTransportError	Base exception
SignatureVerificationError	Signature did not verify
HandshakeError	Handshake failed
SessionExpiredError	Session timed out
ReplayAttackError	Nonce reused
PeerNotAuthenticatedError	No handshake completed

Examples

See the examples/ directory:

• simple_server.py -- Run a PQC MCP server with signed responses
• simple_client.py -- Connect to a server with PQC handshake
• mutual_auth.py -- In-memory mutual authentication demo

Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Lint
ruff check src/ tests/

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/my-feature)
3. Write tests for your changes
4. Ensure all tests pass (pytest)
5. Submit a pull request

License

Apache License 2.0. See LICENSE for details.