Ask a model to compute business health from five data sources and it will improvise a method, differently each time. CorpusIQ's answer is to make the method a stored artefact — a runbook with an id — and execute it server-side, so the same question gives the same answer for the same reason. The tool descriptions enforce the pattern, telling the client which call must come first. The cost is 126 tool definitions; the discovery-first design is presumably why that is bearable.
CorpusIQ's endpoint, which listed 126 tools to our probe under the name multi-source-mcp. The design is unusual: rather than exposing raw data tools, it routes questions to server-side runbooks that execute deterministically.
- resolve_connector — the documented first call for any data question: work out how it can be answered
- select_runbook — for broad business questions, pick the runbook that fits before running anything
- list_runbooks and run_runbook — browse the catalogue, then execute one by id with arguments and a date range
- ask_corpusiq — a business question answered through the deterministic runbook executor
- get_connector_status — a formatted dashboard of every configured connector
- And the rest of the 126, covering the individual connected sources
The handshake and listing answered without credentials; the connectors themselves are what hold your data, and those are configured on CorpusIQ's side.
Paste a URL — add the endpoint to your client