Anthropic's Model Context Protocol has crossed 97 million monthly SDK downloads. Thousands of MCP servers exist — GitHub, Postgres, Slack, web search, file systems. The tool ecosystem is massive.
But MCP connects agents to tools. It does not connect agents to agents.
An MCP-equipped agent can query any database, call any API, read any file. Delivering those results to a peer agent still requires building your own transport, discovery, and encryption. Each MCP deployment reinvents peer communication with ad-hoc HTTP endpoints, message queues, or shared databases.
Think of agent infrastructure as two axes:
Vertical (MCP). Agent reaches down into systems. Database queries, API calls, file access. The agent talks to tools.
Horizontal (Pilot Protocol). Agent reaches across to peers. Discovery, trust, encrypted tunnels, task delegation, pub/sub events. The agent talks to agents.
MCP gives an agent eyes and hands. Pilot gives it a voice and ears. A fully capable agent needs all four.
The integration runs in a single process with two clients: the MCP client connects to tool servers, the Pilot driver connects to the local daemon. Your application logic coordinates between them:
- Receive a research task from a peer via Pilot's task system
- Use MCP to query a database
- Use MCP to search the web
- Produce a summary with your LLM
- Return results to the requester via Pilot
- Publish a
research.completedevent so other interested agents are notified
The two clients are independent — different protocols, different endpoints, different lifecycles. Upgrade your tools without changing your network layer.
A Pilot address gives any MCP agent an identity on a peer network. It can be discovered (or stay invisible). It can receive tasks, publish events, exchange data — all over encrypted tunnels that traverse NAT automatically. The tools still work. The LLM still works. You just added collaboration.
Read more: MCP + Pilot: Give Your Agent Tools AND a Network · Build a Multi-Agent Network in 5 Minutes
Top comments (0)