1. Connect and discover tools
Use the URL shown in Settings → MCP → Connect in your IQRush environment. The remote connection uses Streamable HTTP and browser authorization. See connection setup. For a custom client, let your MCP library handle the protocol lifecycle, then calltools/list. Use its inputSchema to validate arguments. Tool names below are IQRush names; a host may display a prefix around them.
2. Find your project
Calllist_projects with an empty arguments object:
projects and count. Find the intended project by its name and workspace, then retain its project_id. Names can repeat; IDs identify the project.
3. Select a completed run
Calllist_runs with that project_id:
status: "COMPLETED". Check data_ready. For grouped or filtered measurements, also check cold_path_ready when returned. Save run_id and the completion date. If the required data is not ready, wait and check the run again before interpreting an empty measurement.
start_session is a convenient alternative for discovery, but its latest-run selection is not a guarantee that the latest run is complete. Inspect the returned run before using it.
4. Check stability
Callget_results_stability with the project and run IDs. Read data.is_converged, data.rank_corr_ok, data.sufficiency_ok, and data.structural_snr. A finished run and a converged measurement are different states. See decision-grade results.
5. Read citation coverage
Replace the two ID placeholders with values returned by discovery:params of an MCP tools/call request, not standalone HTTP request bodies. See custom clients for the surrounding protocol message.
What success looks like
You receive the requested project and run, rows for the owned domains, platform scope, and anestimate for each row. Display the value with its returned confidence interval. A coverage value of 0.25 is 25%, not 0.25%.
Check errors and warnings even when the transport succeeds. An empty list is not automatically zero visibility. Follow response handling if the result has no rows.
Next: Work through a reporting example, then choose a report recipe. For your own client, use response handling before turning returned fields into a report.