Practical MCP cookbook
These recipes describe complete research jobs rather than isolated tool calls. Let the connected client inspect the live tool schemas before constructing arguments, and confirm a proposed study plan before execution.
Test positioning across two buyer segments
Use this when a positioning decision depends on different buyer roles.
- Use
list_groupsto check whether the audiences already exist. - If needed, use
create_group_from_briefwith a specific population description for each segment. - Create one panel that contains both groups.
- Ask the same focused positioning question across the panel.
- Use
get_panel_summaryto inspect agreement, disagreement, and segment differences. - Export the result for the decision record.
Example population briefs:
Procurement leaders at European manufacturers who buy industrial software.
Operations leaders at the same companies who use the software day to day.
Keep the question neutral. Ask what is clear, credible, relevant, and missing before asking respondents to choose a winner.
Screen a concept before fieldwork
Use a synthetic panel to identify what representative human research should test, not to claim population-level preference.
- Create or reuse the intended audience.
- Create a panel around the decision the concept must inform.
- Ask one diagnostic question first: what does the concept appear to promise?
- Follow with questions about relevance, credibility, objections, and missing information.
- Inspect minority views as well as the majority pattern.
- Convert unresolved assumptions into a human-research brief.
For multiple questions, use plan_panel_study, inspect the plan, then run the
confirmed version with run_panel_study.
Run a durable message study
Use this sequence when the study must survive the initiating chat session.
- Call
plan_panel_studywith the decision, audience, messages, and required comparison. - Review the proposed order and remove leading or redundant questions.
- Confirm the plan.
- Call
run_panel_study. - Poll
get_panel_studyuntil the run completes. - Use
get_panel_summaryfor the whole-study view. - Export the result in the format required by the next consumer.
Durable execution means the server-side run continues independently of the chat session. Preserve the audience definition, approved plan, and export together.
Check method availability before planning
Do not assume that a named research method is executable.
- Call
list_research_methods. - Check the current execution status.
- Select only an available method.
- Use a structured study plan when the method requires multiple tasks.
MaxDiff currently executes through a deterministic adapter. Conjoint is planned but cannot currently execute. The live server response remains authoritative.
Finish with the right validation
Before using a result in a consequential decision:
- inspect segment differences and uncertainty;
- separate evidence-backed findings from plausible hypotheses;
- identify claims that require representative human validation;
- retain the audience definition and study plan;
- export the result into the decision workflow.