Looker MCP Integration

Weldable's Looker MCP integration connects your AI agents to your Looker instance, giving them access to your semantic data layer through natural language. Your agent can run queries against LookML models, retrieve saved Looks and dashboard data, and create visualizations without anyone writing SQL or navigating the Looker UI.

Google introduced the official Looker MCP server as part of the MCP Toolbox for Databases, positioning Looker's semantic layer as the "brain" that AI agents query for trusted business data. Weldable integrates with this approach, handling authentication and intent mapping so your agent can ask questions about your data and get answers grounded in your organization's defined metrics and dimensions.

Use cases

Natural language data queries

Your agent translates plain English questions into Looker API calls against your LookML explores. Ask "what was revenue by segment last quarter?" and the agent selects the right explore, dimensions, measures, and filters, then executes the query in real time through Looker's Run Inline Query API. The results come back formatted and ready to share. Business users get answers without writing LookML or SQL.

Scheduled dashboard digests

Your agent pulls data from key Looker dashboards at a set cadence and compiles the results into a summary message. It compares current metrics against the previous period, highlights significant changes, and posts the digest to your team's Slack channel every Monday morning. Stakeholders get the numbers they care about without opening Looker.

Ad hoc reporting for stakeholders

When an executive asks "how did the new pricing tier perform in APAC this month?" your agent queries the relevant Looker explore, applies the geographic and time filters, and returns the answer within seconds. It can also create a new Look in Looker to save the query for future reference. This removes the turnaround time between a question being asked and a data team member running the report.

Data quality monitoring

Your agent runs scheduled queries against your Looker models to check for anomalies: null values in fields that should never be null, counts that drop to zero unexpectedly, or metrics that deviate more than two standard deviations from their 30-day average. When an anomaly is detected, the agent posts an alert to the data engineering channel with the specific query results and affected dimensions.

Cross-tool analytics workflows

Your agent pulls data from Looker, combines it with campaign data from Google Ads or product data from your application database, and produces a unified analysis. Looker's semantic layer ensures the business logic is consistent (revenue means the same thing everywhere), while Weldable handles the data assembly across sources.

How it works

Connect your Looker instance by providing API credentials through Weldable's setup flow. Looker uses API3 client credentials (client ID and client secret) for authentication. Weldable manages the session token lifecycle and handles re-authentication automatically.

Once connected, describe your question in natural language. Weldable maps your intent to the appropriate Looker API calls, whether that's running an inline query against a LookML explore, fetching results from a saved Look, or pulling tile data from a dashboard. The agent resolves model and explore names, applies filters, and returns structured results.

Tips

LookML models define what your agent can query. Your agent's analytical capabilities are bounded by the explores, dimensions, and measures defined in your LookML project. If a question can't be answered, it's often because the relevant field hasn't been modeled. Work with your Looker developer to ensure key metrics are exposed.

Saved Looks are faster than inline queries. If your agent frequently runs the same query, save it as a Look in Looker. Fetching results from a saved Look is faster and lighter on the Looker instance than re-running an inline query each time.

Looker's API uses model and explore names, not display titles. When configuring workflows, use the programmatic names from your LookML project (e.g., order_items not "Order Items"). Your agent handles this mapping for conversational requests, but explicit references should use the API names.

Row limits default to 500. Looker's Run Inline Query API returns a maximum of 500 rows by default. If your agent needs more, it should specify a higher limit parameter in the query, up to the instance's configured maximum. For very large exports, consider using Looker's async query endpoint.

Permissions follow the authenticated user's access. Your agent can only query explores and data that the API credential's associated user has permission to access. Scope the API user to the models and data your agent needs, and avoid using an admin-level credential for routine queries.