Documentation

The Strickle MCP server

Connect an AI client to one Strickle workspace and it can read your catalogued schema, run read-only queries, and answer from the metric definitions your team has reviewed.

Server URL

https://demo.strickle.ai/mcp

Paste this into any MCP client that speaks streamable HTTP. There is no token to paste — the client signs you in the first time it connects.

Claude.ai and Claude Desktop

Add a custom connector with the server URL above and sign in when prompted.

Claude Code

Add Strickle, then run /mcp in Claude Code to sign in.

claude mcp add --transport http strickle https://demo.strickle.ai/mcp

Signing in

Strickle is its own OAuth 2.1 authorization server, with PKCE. You approve the connection once, for one workspace, and Strickle issues the client a short-lived token of its own — your database password is never shared with it. Revoking a client under Authorized AI clients ends its access immediately.

What a connected client can do

Strickle never writes to your connected database. Every query runs inside a read-only transaction, and anything that is not a single read-only SELECT against a catalogued table is rejected before it reaches the database.

Read

Your catalogued schema with its sensitivity flags, read-only queries against the connected database, your team's reviewed metric definitions, and the recorded provenance of any earlier run.

Write, inside Strickle only

Propose a metric definition for your team to review, and create a dashboard. Both are recorded against you, and a client can neither approve its own proposal nor share anything — those stay decisions someone makes in the Strickle app.

The tools

What a connected client can call. Two of them write, and only inside Strickle.

get_schema: Returns the catalogued tables and columns with types, keys, relationships, sensitivity flags and advisories.
run_sql: Runs a single read-only SELECT against catalogued tables and returns the rows with the exact SQL executed.
search_metrics: Finds the governed metrics that match a question, so a client can prefer a reviewed definition over ad-hoc SQL.
query_metric: Answers from a governed definition, returning the compiled SQL and the lineage behind the number.
get_query_run: Retrieves the stored provenance of an earlier run — its SQL, the definitions it used, and when it ran.
propose_definition: Submits a metric, dimension or filter for the team to review. Proposals carry the proposer's name. Writes inside Strickle
save_dashboard: Saves a named set of blocks whose numbers recompute live. It compiles without running, and mints no share links. Writes inside Strickle