User research through MCP: bring evidence into your agent workflow

What MCP adds to a research workflow

The Model Context Protocol (MCP) connects compatible AI applications to external systems. In a research workflow, the useful question is what the connected research platform makes available: whether an agent can retrieve relevant insight, help prepare a study, or take an action that starts collection.

Palette’s MCP product page presents research access from tools such as Claude Code, Codex, and Cursor. Client support and configuration can change, so use the Palette MCP documentation for the current setup rather than treating a blog example as installation instructions.

The connection does not change the standard of evidence. An agent should be able to distinguish an existing user response from its own inference, and a reviewer should be able to inspect the sources. The goal is to bring research into the workflow with its context intact.

Start by asking what evidence already exists

Before requesting a new study, ask the agent to look for material relevant to the question. A useful request specifies the audience, experience, and uncertainty. For example: “Find evidence about first-time workspace administrators deciding what access to give a teammate. Include relevant sources and explain where the evidence is incomplete.”

That request is more reviewable than “Tell me what users want.” It gives the agent a boundary and asks it to expose gaps. Inspect whether returned material concerns the intended audience, product version, and task. Several mentions of permissions may be less relevant than a single detailed account of the exact decision you are investigating.

If the retrieved evidence is insufficient, say so explicitly. A missing answer is useful information. It can justify a new study or a different method instead of encouraging the agent to fill the gap with an explanation that sounds plausible.

Turn the missing evidence into a study brief

Write the objective before asking for questions. A hypothetical brief might be: “Draft a user study about how recent workspace administrators chose access for their first teammate. Ask about a specific invitation they sent, what they checked, and any information they needed. Return the draft for review.”

Add the audience and the type of evidence you need. If the question is whether people can complete an invitation task on a product branch, request a usability test instead. If it concerns a short question on an existing settings page, a microstudy may fit. The agent should not have to guess whether your uncertainty is about a past experience or a task to attempt.

The research method comparison can help you make that choice. For question wording, use how to write user research questions. A well-bounded brief is useful whether the study is created inside Palette or through a connected tool.

Review the draft and launch details

Preparing a study and starting collection are distinct steps in the workflow. Before launch, inspect the questions, eligible audience, recruitment source, target, and any applicable reward or cost. Confirm the task environment for usability research and the page context for a microstudy. A good draft can still target the wrong people.

Before asking an agent to launch a study, have it present the launch details for your approval. Use that as a review pattern: ask for the proposed setup, resolve ambiguity, and make the launch instruction explicit. The review should be concrete enough that you can tell what participants will encounter.

1. Ask for a draft tied to a clear objective.
2. Review wording, method, and audience.
3. Check the participant experience and collection details.
4. Approve a specific launch when the setup is ready.
5. Check the resulting study status in Palette.

The exact actions available depend on the connected client and your Palette access. Check the product documentation when configuring the connection. Avoid copying secrets into a prompt or a shared example; use the client’s documented configuration path for credentials.

Bring results back with their evidence

When responses are available, ask the agent to answer the original research question and identify support, exceptions, and remaining gaps. “Summarize the findings” is a start. “Explain what we learned about how new administrators choose access, with source responses and contrary cases” gives a clearer analytical task.

Review the original answers behind consequential claims. Make sure the agent has not combined a participant’s account with an implementation suggestion and presented both as a research finding. Record your team’s decision separately so that a future reviewer can see where evidence ended and judgment began.

The AI research synthesis guide describes this review process in more detail. Research access through MCP is most useful when it helps you preserve those distinctions while working in the tool where the question first arose.

Try one question from your current work

Choose a question you are already discussing with your team. Query existing insight, inspect its sources, and decide whether a study is needed. If it is, prepare a draft and review it in the same deliberate way you would in the research platform. Start with a task small enough that you can evaluate the complete path from question to response.

Explore Palette MCP and follow the setup documentation for your compatible client. For the broader approach, read what agentic user research means in practice. The benefit comes from making user evidence available when the question matters, with enough context for someone to judge it.