The shared memory
for your AI swarm.
A local coordination server that holds all task state, agent mailboxes, and knowledge. Every surface — IDE, CLI, voice — speaks the same protocol and sees the same board.
How MeshMCP works
MeshMCP is the single source of truth for every agent in the swarm. It runs locally, starts automatically, and requires zero configuration.
Zero configuration — starts with MeshSpace
When MeshSpace launches, its Tauri backend automatically spawns node server.js as a sidecar process on port 3456. A cryptographic auth key is generated using rand::thread_rng().fill_bytes() and stored in the Tauri secure store.
A Tokio watchdog task polls the server every 20 seconds and restarts it automatically if it crashes. MeshSpace writes the server URL and auth key to .meshpilot/mcp.json in your project directory so every Claude agent process can read it without any manual setup.
Structured task lifecycle with deduplication
Every task moves through a defined lifecycle: todo → in_progress → in_review → done. Agents call update_task to advance their tasks. The Reviewer can reject a task back to todo with a reason, which the Builder reads from the task description on its next poll.
Task creation is deduplicated using a 120-character fingerprint of the title — if the Coordinator tries to create the same task twice, the second call is silently ignored. All writes are serialized through a _writeQueue promise chain to prevent race conditions when multiple agents write simultaneously.
SSE stream keeps the board live
MeshSpace subscribes to GET /events on startup and keeps the connection open. Every time an agent calls update_task or create_task, MeshMCP persists the change to SQLite and immediately broadcasts a Server-Sent Event to all connected consumers.
MeshSpace's boardStore receives the event and re-renders the Kanban column — no polling, no refresh. If the SSE connection drops, MeshSpace reconnects automatically with exponential backoff.
task.creatednew task added to boardtask.updatedstatus or fields changedtask.deletedtask removedchat.messageagent sent a messageproject.eventswarm lifecycle signalThe full API surface
MeshMCP exposes both a REST API and an MCP tool interface so any agent — inside MeshSpace or external — can interact with the task board.
REST API
MCP tool interface
Security
Every request is validated against a cryptographic Bearer key generated at startup. The key is stored in the Tauri secure store and written to .meshpilot/mcp.json for agent processes to read.
SQLite storage
Tasks, chat messages, knowledge, and settings are stored in a local SQLite database. Writes are serialized via a _writeQueue promise chain. normalizeTask() runs on every read to ensure consistent shape.
Deduplication
Task creation is deduplicated using a 120-character fingerprint of the title. If the Coordinator creates the same task twice (common in long swarms), the second call is silently ignored.
External agent support
Any MCP-compatible client can connect to MeshMCP — Claude Code, Cursor, Windsurf, or your own scripts. The same task board that MeshSpace shows is accessible to any tool that speaks the protocol.
MeshMCP is the backbone of every swarm.
It starts automatically when you launch MeshSpace. No setup required.