AssetLab MCP Server

Talk to your assets. Connect Claude, ChatGPT, Microsoft Copilot, or any MCP-compatible client to your AssetLab account through the Model Context Protocol - ask about work orders, PM schedules, capital plans, and infrastructure networks in plain language, and create or update records hands-free.

480+ tools across 100+ resources · scoped, tenant-bound, audited · read and write · hosted or local

Quick start

Claude.ai (hosted - easiest)

1. Create an API key in AssetLab → Settings → API Keys
2. In Claude.ai: Settings → Connectors → Add connector
3. Paste the connector URL: https://mcp.assetlab.ca
4. When prompted for auth, paste your API key (al_live_...)

ChatGPT

1. Create an API key in AssetLab → Settings → API Keys
2. In ChatGPT: Settings → Apps & Connectors → Add new connector
3. Name it (e.g. AssetLab) and enter the server URL: https://mcp.assetlab.ca
4. Set authentication to OAuth and click Create - the OAuth endpoints are auto-discovered
5. On the AssetLab authorization page, paste your API key (al_live_...)

Microsoft Copilot (via Copilot Studio)

Copilot chat has no "add a server" field, so one person builds an agent once and publishes it; staff then use that agent inside ordinary Microsoft 365 Copilot.

1. Create an API key in AssetLab → Settings → API Keys
2. In Copilot Studio, open an agent with generative orchestration and choose Tools → Add a tool → New tool → Model Context Protocol
3. Server URL: https://mcp.assetlab.ca/mcp?profile=core
4. Authentication API key, type Header, header name Authorization; the connection value is Bearer al_live_...
5. Publish the agent to the Microsoft 365 Copilot channel and have an admin approve it

Use the ?profile=core address here. Copilot Studio caps an agent at 128 tools and recommends 25-30; this server publishes 480+, and the core profile answers with 28 covering sites, buildings, assets, work orders, work requests, PM schedules and vendors. Claude and ChatGPT have no such cap and should use the plain URL.

Claude Desktop / Claude Code (local, stdio)

This package runs as a local MCP server over stdio - nothing is exposed on the network.

{
  "mcpServers": {
    "assetlab": {
      "command": "npx",
      "args": ["-y", "@assetlab/mcp-server"],
      "env": {
        "ASSETLAB_API_KEY": "al_live_...",
        "ASSETLAB_API_URL": "https://<your-project>.supabase.co/functions/v1/api-gateway"
      }
    }
  }
}

Your AssetLab administrator can provide the API URL for your organization.

What can you ask?

What's new in 2.7

A curated tool profile for Microsoft Copilot. Adding ?profile=core to the hosted server URL publishes 28 tools instead of 480+, which is what makes the server usable in clients that cap how many tools one agent may host - Microsoft Copilot Studio allows 128 and recommends 25-30. An unrecognized profile name is rejected rather than quietly serving the full catalog. Claude and ChatGPT are unaffected and keep the full tool set.

What's new in 2.5

Facility lifecycle strategies. The lifecycle strategy tools now cover buildings and equipment, not just linear infrastructure. Five new tools under the asset_lifecycle_events scope read and maintain condition-triggered interventions ("roof recoat at condition 70-85 adds five years"). Events attach to an asset type or an asset type group rather than to individual assets, so one event maintains a whole class of equipment.

Events say what work they become. Lifecycle events on both sides now carry a work_generation setting - plan only, work order, or project - so a due intervention records what it should turn into. Nothing is created on a schedule; the interventions lists remain where that decision is made.

Tool catalog

Tools follow one naming pattern throughout: list_* / get_* to read, create_* / update_* / delete_* to write. Read below means list/get; Write means create/update/delete (availability varies slightly per resource - the tool descriptions are the authority).

Scopes

API keys use scopes to control exactly what your assistant can touch. Configure per-resource read and write access in Settings → API Keys - scopes are grouped by category (Operations, Facilities, Projects, Finance, Infrastructure, ...) with bulk-toggle controls.

Security

• OAuth 2.0 + PKCE through mcp.assetlab.ca; your API key is encrypted in transit and never stored by the MCP server
• Tenant-bound - every key is scoped to your organization only, and the server ignores any attempt to address another tenant
• Audited - all data access is logged in AssetLab's audit log
• Revocable - revoke a key any time in Settings → API Keys; it dies instantly
• Read-only by default - write operations require explicit write scopes

Support

Questions or issues? Email [email protected].

Docs: assetlab.ca → Docs → AI & MCP