AI Assistants · Model Context Protocol
Click-driven dashboards were built for finding information. Prompts are built for getting it. Bringup™ now speaks MCP — so Claude Code, Codex, or any MCP client you already use can read your fleet, logs, and org directly, and answer in plain language.
read-only · loopback-only · opt-in
The Bringup workbench is still there for deep work. But the questions you ask twenty times a day should not cost twenty navigation paths.
Click into the fleet panel, filter, open a device, cross-check the logs view, then the org page. Every answer costs a navigation path you have to remember.
Ask "which devices are offline and why" and let the assistant walk the same data — devices, logs, versions, roles — and hand back one synthesized answer.
The question comes up mid-debug, in your terminal or editor. Now the answer arrives there too — no context switch to a dashboard and back.
Your assistant composes these from live tools — no stale exports, no copy-pasted log files.
Which devices in my fleet are offline right now?
What daemon version is warehouse-agv-01 running, and does it need an upgrade?
Show the last 50 error-level logs from the fleet extension.
Who has admin access in my organization?
Is my local daemon healthy and enrolled?
Summarize what changed across the fleet since yesterday.
Extensions contribute tools through the same manifest point the core uses — install an extension, and every connected assistant learns its tools automatically.
bringup_statusCoreApp version, signed-in user, active organization and profile, local daemon health — the orienting call every session starts with.
logs_queryCoreApp, daemon, and per-extension logs with level, source, and text filters. Debugging without leaving the conversation.
fleet_list_devicesFleet extensionEvery device with status, type, daemon version, and role — the "what is my fleet doing" question, answered in one call.
fleet_get_deviceFleet extensionOne device in full detail: lifecycle, groups, labels, addresses, last seen.
org_currentOrg extensionThe active organization, your role in it, and every org you belong to.
org_list_membersOrg extensionMembers, roles, and pending invitations for the active organization.
one manifest entry · every assistant updated · live in settings
Turn on AI assistant access in the desktop app (Settings → AI Assistants), then register Bringup in your client. The bridge reads the port and token automatically — no secrets land in any config file.
$ claude mcp add bringup -- npx -y @bringuplabs/bringup-mcp-server# requires Node.js — connection details are discovered automatically
[mcp_servers.bringup] command = "npx" args = ["-y", "@bringuplabs/bringup-mcp-server"]
{
"mcpServers": {
"bringup": {
"command": "npx",
"args": ["-y", "@bringuplabs/bringup-mcp-server"]
}
}
}Clients that speak streamable HTTP natively can connect directly — the app's Settings page shows the endpoint and a per-user token, plus these commands with copy buttons.
Giving a language model eyes on your fleet only works if the blast radius is engineered to zero. It is.
MCP access is an explicit toggle in Settings. Every client connection surfaces in the notification center — nothing talks to your fleet silently.
The endpoint binds to 127.0.0.1 with a bearer token stored owner-readable-only, plus DNS-rebinding protection. Your network never sees it.
Version one answers questions; it does not push buttons. No deploys, no deletes, no shell commands — actions arrive later, behind per-action approval.
A central redaction layer strips tokens, keys, and certificate material from every tool result before it leaves the app.
Ships with the Bringup desktop app. Flip the toggle, paste one command, and ask your first question.