AI agents can write a research plan in seconds. The hard part begins one step later: turning that plan into real conversations with customers, tracking the work, and bringing the evidence back without losing context along the way.
Maren MCP closes that gap. It lets an approved agent use Maren as a research capability: inspect a workspace, choose an interview method, create a project, prepare participant links, monitor completions, and retrieve cross-interview synthesis.
The customer remains in charge of the important boundaries. You choose the workspace. You review the permissions. Adding a participant does not send a message. Maren still enforces research methods, access rules, and usage limits. The agent helps operate the workflow; it does not become the system of record.
The short version
Maren MCP turns customer research into a structured set of agent tools, so a conversation about a product decision can become a reviewable study and, later, traceable findings.
What MCP changes
The Model Context Protocol gives AI clients a standard way to discover and call tools exposed by another product. In practical terms, it means Maren can describe what an agent may do, what information each action needs, and what result comes back.
Without that connection, your agent can help write a discussion guide, but you still have to copy it into a research tool, configure the project, add participants, check progress, run the analysis, and carry the result back. Every handoff is a chance to lose the question you were originally trying to answer.
With Maren connected, the agent can work with the same context that produced the research need. If a planning conversation reveals uncertainty about trial activation, you can ask the agent to prepare a study about that uncertainty. It can inspect the available interview styles and workspace capacity before creating anything.
Ten tools cover the research workflow
The first Maren MCP release exposes ten focused tools. They fall into five stages:
| Stage | What the agent can do |
|---|---|
| Orient | Check the connected workspace, current capacity, available interview methods, and existing projects. |
| Design | Create a draft with an objective, interview style, depth, language, tone, and discussion guide. |
| Prepare | Add participants, activate the project, and retrieve the unique interview link that belongs to each person. |
| Monitor | Read a compact snapshot of participant states, interview completions, synthesis readiness, and the next useful action. |
| Learn | Start cross-interview synthesis once at least three eligible interviews are complete, then retrieve the report. |
These are not screen-automation shortcuts. Each tool uses structured inputs and results. Maren validates the request and applies the same project, quota, and workspace rules used by the product.
A research workflow can begin with a sentence
Imagine a product conversation that ends with this:
We keep debating why trial users do not invite a teammate. Prepare interviews that can tell us what happens between first value and collaboration.
An agent with Maren access can turn that into a concrete sequence:
- Check how many additional interview starts are currently available.
- Compare Problem Discovery with UX Discovery and explain the trade-off.
- Prepare a draft project and discussion guide for review.
- Add the participant emails you provide, without contacting anyone.
- Activate the approved project and return the matching interview links.
- Check progress later and tell you when there is enough evidence to synthesise.
- Bring the completed synthesis back to the planning conversation.
The time saved is useful. The continuity is more important. The research question, project design, fieldwork status, and findings can stay attached to the decision that created the study.
What agents gain
A real path from reasoning to evidence. An agent can notice a gap in what the team knows and help turn it into a study, instead of filling the gap with a confident guess.
Method information it can act on. Maren returns the purpose, methodology, supported depths, typical duration, and example questions for each interview style. The agent can recommend a method based on the decision at hand.
Clear next actions. Project status does more than return counts. It indicates whether the useful next step is adding participants, activating the project, waiting for interviews, starting synthesis, or retrieving the result.
Business rules at the tool boundary. The agent does not have to infer whether a project is ready or whether capacity remains. Maren answers those questions and rejects invalid transitions.
What customers gain
Less translation work. The goal and constraints already discussed with an agent can become the first draft of the study. You review the work instead of rebuilding it.
A repeatable research habit. Teams can use the same sequence for activation questions, churn, feature feedback, win/loss research, and other recurring decisions. The agent handles the checklist; Maren handles the interviews and synthesis.
One place to revoke access. Connections appear in Maren settings with their status and last-used time. Disconnecting one blocks its next request.
Human control where it matters. A person still decides what the study should answer, which participants belong in it, whether links should be distributed, and what the findings mean for the business.
The guardrails are part of the product
Giving an agent one broad API key would be simple. It would also hide the most important questions: Which workspace? Which actions? Which customer data? Who approved this connection?
Maren uses OAuth instead. During connection, you select one workspace and review the requested permissions. The server derives the user, client, workspace, and scopes from that approved connection; the agent cannot choose another workspace in a tool call. Maren checks active membership again on every request.
The individual permissions separate reading projects, creating research, managing participants, accessing participant links, and running synthesis. Sensitive participant links require their own permission and are treated as credentials.
There are deliberate limits too:
- Adding participants never sends invitations, reminders, or thank-you emails.
- Creating or activating a project does not silently consume an interview start.
- Potential paid overage requires explicit confirmation before activation.
- Every dispatched tool operation receives an audit reference.
- The first release returns synthesis, not raw interview transcripts.
- Participant-supplied content in synthesis is marked as untrusted research data, not agent instructions.
Those boundaries are not friction added after the useful part. They are what makes the useful part suitable for real customer research.
What Maren MCP does not do
MCP does not turn Maren into an autonomous research department. It does not choose the business decision, recruit a representative panel, decide whose consent is sufficient, distribute participant links, or make the product decision after the synthesis arrives.
It also does not replace Maren’s participant experience. The adaptive interview still happens in Maren, using the selected research method. The synthesis still comes from Maren’s cross-interview analysis. MCP makes those capabilities available inside an agent workflow.
That division of labour is intentional: let the agent carry context and coordinate the steps; let Maren run the research; keep people responsible for purpose, contact, and judgment.
Connect Maren to your agent
Maren provides guided setup for Claude and Codex. In Maren, open Settings → AI agents, copy the connection URL, and follow the instructions for your client. The sign-in flow asks you to select a workspace and approve access. No shared API key or environment variable is required.
Once connected, start with a bounded request: ask the agent to inspect the workspace and list interview styles. Then have it prepare a draft study for review before adding participants or activating anything.
See everything Maren MCP offers, including the ten tools, customer safeguards, and answers to common connection questions.