Skip to main content
The Less MCP server lets AI agents - like Claude, ChatGPT, and Cursor - work directly with your Less workspace. An agent can find your sources, models, and tables, answer data questions with SQL, trace lineage, and check job health, all through a single secure endpoint. It speaks the Model Context Protocol and uses OAuth 2.1 with automatic discovery (RFC 9728 Protected Resource Metadata, Client ID Metadata Documents, and Dynamic Client Registration), so MCP clients can sign you in and let you select your workspace right in your browser without manually copying API keys.

Endpoint

The default workspace host is app.less.tech. If your workspace uses a different host, use that instead.

Multiple workspaces

If your Less account belongs to more than one workspace, you can connect them in two ways:
Add a separate MCP connector for each workspace using a workspace-specific URL:
<workspace-id-or-name> can be the workspace UUID or a slug of its name (lowercase, spaces become -). If two workspace names share a slug, the pinned URL is rejected and you need the UUID. Each connector stays pinned to that workspace, which is useful when you want Claude or ChatGPT to treat two companies as two distinct tools.
Every call runs as you in the workspace that is currently active, and is limited to what your account may read or execute. The agent only ever sees assets you have access to in that workspace.
Workspace admins can see MCP activity in Settings → Audit log, where asset runs and data reads made through the server are recorded with the origin MCP.

What the agent can do

The server exposes a focused, read-mostly set of tools:
  • Discover - search assets, open an asset’s details, list tables and their columns.
  • Analyze - preview sample rows and run aggregate SQL over verified tables.
  • Understand pipelines - trace upstream provenance and downstream impact, walk a model step by step, and find where a metric is computed.
  • Operate - review schedules, check daily job health, and read a run’s logs to diagnose failures.

Running assets

Reading and analyzing never changes anything. The one tool that does is executeAsset, which starts a job to refresh a source, model, orchestration, or Python script. Agents are instructed to call it only when you explicitly ask to run or refresh something, and to check that you have permission first.

Get connected

1

Add the Less MCP URL to your agent

Follow the guide for your client: Claude Desktop or ChatGPT desktop.
2

Sign in & choose your workspaces

Your client will open a browser window to Less where you can sign in, tick the workspaces the agent may use, pick a default, and click Allow access.