Connect AI Tools to Raw Research Data and Save Results to Condens with MCP

Many researchers and product teams like to work in Claude, ChatGPT or Copilot, right next to everything else on their plate. What those tools produce depends on what they're given. As Condens CEO Alexander Knoll often puts it, AI-supported insights are only as good as the context they draw from: the problem behind what people said, the metadata around it, and what your organization already knows.

A transcript pasted into a chat carries none of that. The answer can sound right and still miss what matters, and the work stays in one person’s chat history, with no link back to the participant.

A research repository holds those layers. With the Condens MCP Server, your AI tools work from all of it and save what they produce back to Condens. You work in the tool that fits the moment, while Condens keeps the evidence and the structure behind it.

The Condens MCP Server lets AI tools and agents like Claude, ChatGPT, Microsoft Copilot Studio and company LLMs read your Sessions, Transcripts and unpublished Artifacts, create Highlights and Notes, and update existing Artifacts in Condens. Admins decide which of these abilities each connection gets. Every ability is its own permission, so you turn on exactly what fits your team.

A connection that reads published Artifacts only is still the right setup for many teams. Stakeholders get curated, reviewed findings, and work in progress stays with the research team. Raw data and write access build on that for the people who work closest to the data, like researchers, ReOps and contributors on product teams. On Enterprise, permission sets let you offer each of these groups its own level of access through the same connection.

The questions that matter most rarely fit inside one study. With raw data access, your AI tool can read the Sessions, transcripts and Session information in every Project you have access to, together with Artifacts that aren't published yet. That makes cross-study work something you can do on a normal timeline, across years of research and next to your other data.

Research is rarely used on its own. A product decision also depends on the roadmap, the backlog, analytics and what sales and support are hearing. The Condens MCP Server works with any MCP-compatible AI tool, so your research can sit next to those sources in the same conversation, with Condens as one of the inputs.

Think of the last product decision you helped with. The roadmap is in Notion, the tickets are in Jira, the numbers are in an analytics dashboard, and the research is in Condens. Inside Condens, AI search is the quickest way to answer a research question. When the question spans research and everything else, being able to ask in the tool you already have open makes a real difference. You stay in your flow, follow-up questions build on the same conversation, and the draft you end up with already has the evidence in it.

The same goes for stakeholders. A product lead who lives in Copilot or a company LLM can get evidence-backed answers there, from exactly the research you've made available to them.

That includes the tools enterprise teams standardize on: Microsoft Copilot Studio, Glean, a company LLM, design tools like Figma Make and coding assistants like Cursor or Claude Code. Research shows up where each team already works, with the same permissions behind it.

The Condens MCP Server also works with agent platforms like n8n, Microsoft Copilot Studio and company LLMs, so research can become part of workflows that run without someone starting each conversation. Agents use the same tools and permissions as everyone else: they read what the connection allows, and can write Highlights, Notes and updates back to Condens.

A few workflows teams are setting up or exploring:

• A weekly research digest: An agent reads the Sessions added in the past week, summarizes what came up by Tag, and shares the summary with your team, with links back to Condens.

• Research combined with product data: An agent pulls findings from Condens together with behavioral analytics and support tickets, and flags where what users say and what they do point in the same direction.

• Stakeholder questions answered where they're asked: A Copilot Studio agent or company LLM answers questions like "what do we know about onboarding?" from your published Artifacts, with links back to Condens, so stakeholders get evidence-backed answers in the tools they already use.

• Findings pulled out of existing reports: An agent goes through older reports and creates Notes for the individual findings, each with its supporting Highlights embedded, so they're easier to search and reuse.

• A research check before new work: Whenever a new study or epic is created, an agent looks for related research in Condens and adds the links, so nobody starts from zero.

For workflows that write to Condens without someone confirming each step, permissions matter even more. A dedicated MCP client with only the abilities that workflow needs keeps it focused, and new Notes still arrive unpublished for a person to review.

The Condens MCP Server can write results back to your repository as Highlights and Notes, so analysis done in an AI tool becomes something the whole team can find and build on.

The Condens MCP Server can update existing Artifacts, so your AI tool can add new findings to a report, rewrite a section or tighten a summary directly in Condens. It works for any Artifact you can edit, except Whiteboards.

This matters most for the reports people keep coming back to: a persona overview, an onboarding report, a running list of usability issues. When two new Sessions come in, you can ask your AI tool to add what's new to the report with the supporting Highlights embedded, and the report stays current without a full rewrite. Comments on the content it changes are kept, so the discussion around a finding stays with it.

Edits are saved right away, including in published Artifacts, so a published report reflects the change as soon as it's made. Your AI tool is instructed to only edit when you ask for it or approve it, and tools like Claude and ChatGPT usually ask you to confirm before an edit runs.

A few things stay in Condens for now. Your AI tool can replace or remove images, files, embeds and charts, but it can't add new ones.

Giving an AI tool access to raw research raises fair questions. Who can see participant data? Could it change a report nobody meant to touch? The Condens MCP Server is set up so you can answer those questions before anyone connects.

• It follows your existing access: The MCP never shows anyone something they couldn't already see in Condens. Viewers can read published Artifacts, and raw data is available to people with a contributor seat.

• Every ability is its own permission: Reading raw data, creating Highlights, creating Notes and updating Artifacts are turned on separately, so a connection set up for reading never starts writing by itself.

• Different teams can get different access: On Enterprise, permission sets let Admins give researchers raw data and write access while stakeholders read published Artifacts only.

• Your privacy settings carry through: Speaker names follow each Artifact's publishing settings, redactions stay redacted, and participant data only appears when the connection has permission to read it.

• You stay in the loop on every change: Your AI tool is instructed to check with you before it saves Highlights or edits an Artifact, and Notes stay unpublished until you publish them in Condens.

The easiest way to start is to paste a Condens link into your AI tool and ask it to work with it. The prompts below give your AI tool clear criteria first, so you get evidence you can check before anything is interpreted or saved.

Research has always been most valuable when it builds up over time. AI tools make it faster to work with, and a structured repository makes sure that work doesn't scatter. With raw data access and write access, the Condens MCP Server brings the two together, so every question your team works through in an AI tool can add to what your repository knows. Condens' own AI features stay right next to your data, for whenever you'd rather work inside Condens.

To set it up, an Admin turns on the permissions you want for each MCP client in Settings > Account > API & MCP-Server. If you're already connected, refresh the connection in your AI tool to pick up the new tools. You'll find step-by-step guides in our Help Center, and if you have questions, reach out at hello@condens.io!