Zed
Zed speaks MCP over stdio. fleet-bridge registers like any other command-based MCP server.
Server registration
Add an entry to the context_servers (or mcp_servers, depending on your
Zed version) map in settings.json:
{
"context_servers": {
"fleet-bridge": {
"source": "custom",
"command": "node",
"args": ["/absolute/path/to/fleet-bridge/dist/index.js"],
"env": {
"FLEET_BRIDGE_AGENTS": "/absolute/path/to/agents.json",
"FLEET_BRIDGE_HERMES": "/absolute/path/to/hermes-agents.json"
}
}
}
}
The key requirement is that command + args launch the built server —
dist/index.js must exist (npm run build). The env map is where roster
paths and timeout overrides go.
Zed's exact configuration keys have shifted across releases (context_servers
vs mcp_servers, source: "custom"); check Zed's current MCP docs for the
map name in your version. The server-side shape (command, args, env) is
identical in all of them.
Using the tools
Once connected, Zed's assistant sees the six tools: list_agents,
agent_health, start_session, invoke_agent, list_hermes_agents,
invoke_hermes. See the tool catalog.
- Progress notifications:
invoke_agentandinvoke_hermesemitnotifications/progresswhen the client passes_meta.progressToken. Clients that do not pass one (or do not render progress) still get the same final result — the notifications are purely additive. - Timeouts: Zed applies its own call timeout on top of the bridge's.
A bridge-side
timeout_mslonger than the client budget means the client cancels first. - Roster reload: restart the server (or the Zed session) after editing
agents.json/hermes-agents.json— rosters load once at startup.
Smoke test
Verify standalone first — see Getting started.