- What is the Condens MCP Server?
- What the MCP can read
- What the MCP can write (Beta)
- Highlights and Tags (Beta)
- Notes (Beta)
- Access permissions
- Permission sets (Enterprise, Beta)
- Connect Condens to Claude and ChatGPT
- Connect Condens to a custom MCP client
- Stage 1: Admin enables the MCP on workspace-level
- Stage 2: Each user authorizes the MCP client for their account
- Creating a dedicated user account to connect with MCP server
- Stage 1: Create a dedicated Viewer account
- Stage 2: Admin configures the MCP client
- Stage 3: Each user authorizes the MCP client
- Refresh your MCP connection after updates
- Redirect URI wildcard support
- Removing or deactivating the MCP client
- Getting started
- Limitations
What is the Condens MCP Server?
MCP (Model Context Protocol) is an open standard that allows AI tools to securely connect to external data sources. Once connected, you can:
Ask questions about your research in natural language and get answers based on your Condens data
Query across multiple Condens workspaces at once
Combine Condens with other MCP-compatible tools (Notion, Jira, Linear, etc.) in the same conversation
Save Highlights, Tags and Notes created in your AI tool back to Condens (Beta)
Writing back to Condens keeps your repository the single source of truth. Analysis done in an AI tool doesn't stay locked in one person's chat or a slide deck. It ends up in Condens, where your whole team can find, reuse and build on it.

About Beta features: Capabilities marked Beta are available to customers in the MCP beta and may still change. To join, contact us at hello@condens.io.
What the MCP can read
The MCP only reads what you already have access to in Condens, and only what the MCP client's access permissions allow.
What it can read | Details | Access permission | Status |
|---|---|---|---|
Published Artifacts | Everything published to your Magazine, across the workspaces you can access | Read published Artifacts | Available |
Unpublished Artifacts | Any Artifact you can access through your Projects. Highlights come through with their full text and Tags, so your AI tool can quote them directly. It can also read charts and expand clusters. | Read Artifacts in Projects | Beta |
Projects and Sessions | Raw research data, including transcripts, notes and Session information fields | Read Projects and Sessions | Beta |
Participant data | Participant profiles from your Participant pool, looked up by Participant ID (for example, from a Session) | Read Participant data | Beta |
Your Tags | Tag names, descriptions, Tag groups and Tag boards, so your AI tool can pick the right Tags when creating Highlights | Create Highlights & assign Tags | Beta |
Speaker names and participant data
The MCP follows each Artifact's publishing settings for showing speaker names.
If an MCP client doesn't have the Read Participant data permission, participant data won't appear in any MCP response.
Redactions you've applied to a transcript carry through, so anything redacted in Condens stays redacted.
Examples
"What did participants say about onboarding in the Q3 interview Project?"
"Summarize this Session and quote the most relevant Highlights: [Session link]"
"Which of our published findings mention pricing concerns?"
What the MCP can write (Beta)
With write permissions, your AI tool can save what it finds back to Condens. That way, analysis done in a chat becomes part of your shared repository instead of staying in one conversation.
What it can write | Details | Access permission | Status |
|---|---|---|---|
Highlights and Tags | Creates Highlights in Sessions and assigns existing Tags | Create Highlights & assign Tags | Beta |
Notes | Creates unpublished Note Artifacts | Create Artifacts | Beta |
Highlights and Tags (Beta)
Your AI tool is instructed to check which Highlights to create and which Tags to use, and to confirm with you before saving. How consistently it asks depends on the AI tool.
The MCP only uses your existing Tags. It can't create new ones.
If a matching Highlight already exists, Condens reuses or extends it and adds the new Tags without replacing existing ones. This prevents duplicates.
After saving a Highlight, your AI tool gets a direct link to open and share it in Condens. New Highlights follow your Project's usual access settings.
On supported MCP clients, the Tag Session in Condens prompt guides your AI tool through analyzing and tagging a complete Session.
Notes (Beta)
Notes are created unpublished, so you can review them in Condens before sharing.
Notes support all text formatting, plus tables, checkboxes, quotes and dividers, and can embed Highlights, Artifacts and clusters.
When a Note is missing details, your AI tool can search your Sessions, Highlights and published Artifacts to fill the gaps. If it finds something relevant in a Session that isn't a Highlight yet, it can suggest creating one.
The MCP can only create Notes. Other Artifact types, like Whiteboards, aren't supported yet.
Editing existing Artifacts isn't possible yet. If you want changes, your AI tool will offer to create a revised version and may suggest deleting the old one. Deleting Artifacts happens in Condens, not through the MCP.

Tip for Claude users: Claude can also create its own artifacts, so a request like "Create an Artifact" may produce one in Claude instead of in Condens. Be specific and say "Create a Condens Artifact" or "Save this as a Note in Condens."
Examples
"Analyze this Session and create Highlights for every mention of pricing. Use our existing Pricing Tags: [Session link]"
"Create a Condens Note summarizing what we know about onboarding drop-off, with the supporting Highlights embedded."
"Find the moments in this Session where the participant talks about exports and save them as Highlights in Condens."
Access permissions
Admins decide what each MCP client can do in Condens. Access always stays within each user's own Condens permissions: the MCP never shows anyone something they couldn't already see in Condens.
To change what an MCP client can access, go to Settings > Account > API & MCP-Server, select the MCP client and choose its access permissions.
New capabilities and defaults: If an MCP client has access to everything in Condens, new capabilities are included automatically when they're released. If you've set specific access permissions, new capabilities stay off until an Admin turns them on.
Write access is separate from read access: Creating Highlights and Notes each need their own permission, so an MCP client set up for reading won't start writing to Condens by itself.
Permission sets (Enterprise, Beta)
On the Enterprise plan, Admins can manage MCP access with permission sets. Each set defines who an MCP client is available for and what it can access, so different teams can get different levels of access. For example, researchers can work with raw data while stakeholders only read published Artifacts.
Existing MCP clients keep their current access as a single permission set until you change it.
If your organization has multiple workspaces, you can choose which workspaces each permission set applies to.
If someone matches more than one permission set in a workspace, they get the combined permissions of all of them. For example, a set for a role and a set for a user group add up. We recommend setting them up so they don't overlap.
Once you have more than one permission set, you can name each one to keep track of who it's for.
To add one, go to Settings > Account > API & MCP-Server, select the MCP client and click Add permission set.
Connect Condens to a custom MCP client
Custom MCP clients in Condens span across a variety of tools you can connect:
Custom or company LLMs
Microsoft Copilot Studio and n8n agents
Enterprise search and AI assistants like Glean
Design assistants like Figma Make
Coding assistants like Claude Code, VS Code, Cursor and Gemini CLI
You can follow the tool-specific setup guides linked above, and for many tools, you will find the presets available once you click on .

From there, setting up a custom MCP client happens in two stages:
A Condens Admin configures the client and sets access permissions for their workspace.
Each authorized user connects the client individually from their own account.
The steps outlined below work for any custom MCP server, including connecting to custom or company LLMs.
Stage 1: Admin enables the MCP on workspace-level
In Condens, an Admin needs to:
Go to Settings > Account > API & MCP-Server and click Add MCP Client.

Enter client details:
Client name: The name that will appear in your authorized apps list
Logo: Upload a logo to help users identify the client easily
Description: Add a short description displayed on the app card (visible to users when browsing authorized apps)
Enter the OAuth2 redirect URI where your service sends users after login. This is required for the authorization flow to work correctly.
For in-house LLMs, consult your development team
For public clients, contact us at hello@condens.io.
If your client uses dynamic redirect URIs, see Redirect URI wildcard support below.

Configure which users, roles, or user groups can use this MCP client, and what it can access (see Access permissions above). Only authorized users will see this client in their Integrations list.
Add setup instructions for your team:
Write step-by-step instructions explaining how to connect to this client.
These instructions appear in Settings > Integrations for authorized users
Tip: Include any necessary URLs, configuration details, or troubleshooting tips.
Click Create.
Once saved, the custom MCP client will appear in the authorized apps list and in Integrations for users who have permission to use it.

Stage 2: Each user authorizes the MCP client for their account
As the final step, each user needs to individually connect the MCP client to their Condens account.
In Settings > Integrations, you'll find the setup instructions provided by the Admin by clicking Show setup instructions. Please bear in mind that the authorization itself happens in the MCP client, not in Condens.

The toggle next to the MCP client in Integrations shows the connection status. If it's off, the authorization step hasn't been completed yet. Once you've authorized Condens in your MCP client, the toggle will automatically turn on.
Creating a dedicated user account to connect with MCP server
Instead of connecting individual user accounts, you can create a single shared Condens viewer account that acts as the MCP connection for your whole company. Everyone using the LLM will see the same Condens content, scoped to whatever that account has access to.
This works well when you want centralized, curated access or when managing individual authorizations isn't practical. Unlike individual integrations, access cannot be managed or scoped on a per-user level; everyone using the LLM sees the same Condens content.
This is done in three stages:
Create a dedicated Viewer account and set its access permissions.
A Condens Admin configures the client and sets access permissions for their workspace.
Then, each authorized user connects the client individually from their own account.
Stage 1: Create a dedicated Viewer account
Create a dedicated Condens user account with a generic email like company-llm@yourcompany.com or research-bot@yourcompany.com
Set the user role to Viewer (this doesn't require a paid seat).
Configure access permissions for this account. Grant access only to the workspaces and projects you want exposed through the LLM.
Use multiple Magazines to create curated views, like a "C-Level Insights" Magazine for executive summaries or product-specific Magazines for different teams. Only publish content you want accessible company-wide.

Stage 2: Admin configures the MCP client
In Condens, an Admin needs to:
Go to Settings > Account > API & MCP-Server and click Add MCP Client (or open the pre-configured Claude or ChatGPT client).

Enter the client details:
Client name: The name that will appear in your authorized apps list
Logo: Upload a logo to help users identify the client easily
Description: Add a short description displayed on the app card (visible to users when browsing authorized apps)

Enter the OAuth2 redirect URI for your service. If your client uses dynamic redirect URIs, see Redirect URI wildcard support below.
In the Available for field, add only the dedicated Viewer account you created in Stage 1.
Add setup instructions for your team:
Write step-by-step instructions explaining how to connect to this client.
These instructions appear in Settings > Integrations for authorized users
Tip: Include any necessary URLs, configuration details, or troubleshooting tips.
Click Create.
Stage 3: Each user authorizes the MCP client
Because all users connect through the shared Viewer account rather than their own, the authorization step works a little differently here.
Each user needs to:
Open an incognito window or log out of their personal Condens account.
Log into Condens using the shared Viewer account credentials.
Go to Settings > Integrations and click Show setup instructions that the Admin prepared for this MCP client. The toggle next to the MCP client in Integrations shows the connection status. If it's off, the authorization step hasn't been completed yet.

Complete the authorization in the MCP client following the instructions provided.
Once authorized, the toggle next to the MCP client in Integrations will turn on automatically. Each user only needs to do this once. This gives your custom or internal LLM access to select data in Condens, regardless of whether users of that LLM have Condens accounts.
Refresh your MCP connection after updates
When we add new MCP capabilities or you change an MCP client's access permissions, existing connections need a refresh to pick up the changes:
ChatGPT: Open the Condens app or plugin settings and refresh its actions.
Claude: Open Customize > Connectors and reconnect Condens.
Claude Code: Run /mcp, select Condens and choose Reconnect.
VS Code: Run "MCP: Reset Cached Tools" from the Command Palette, then use "MCP: List Servers" to restart Condens.
Microsoft Copilot Studio: Open the agent's Tools and edit the Condens MCP server entry to refresh the connection and tool list.
If you don't see a refresh option, disconnecting and reconnecting Condens works too, but you may need to authorize Condens again.
Redirect URI wildcard support
Redirect URIs may include a single trailing wildcard (*). This allows clients with dynamic redirect URIs (like Copilot Studio) to be authorized using a common prefix. The wildcard must appear only at the end of the URL and matches any characters after the prefix.
Example pattern: https://custom-llm-client.app/connector/*
This would allow redirect URIs such as:
https://custom-llm-client.app/connector/a891k2
https://custom-llm-client.app/connector/b1271k
Working with dynamic localhost ports
If you're connecting a local MCP client where the port number changes on each run (common in local development setups), use the localhost wildcard as your redirect URI:
http://localhost:*http://127.0.0.1:*
And if you also need to include localhost without a port number, we recommend adding:
http://localhost/
http://localhost
http://127.0.0.1/
http://127.0.0.1
This matches any localhost address regardless of port, so you don't need to update the redirect URI each time the port changes.

Check the MCP servers directory for more available integrations.
Removing or deactivating the MCP client
If an Admin removes or deactivates an app from the authorized apps list, that app will stop working for all users. Likewise, changing access permissions will limit which users can continue using the MCP server.
Getting started
The easiest way to start is to paste a Condens link, like a Session or an Artifact, into your AI tool and ask it to work with it. For example: "Summarize the key pain points in this Session and suggest Highlights for them: [Session link]".
If you plan to save results back to Condens, mention Condens by name in your request, so your AI tool knows where the result should go.
Limitations
You only see workspaces, Projects and Artifacts you already have access to in Condens.
Without read access to Projects and Sessions (Beta), the MCP can only read published Artifacts.
Without write permissions, AI can't create, edit or delete content in Condens.
With write permissions (Beta), AI can create Highlights, assign existing Tags and create Notes. It can't create new Tags, edit or delete existing Artifacts, or create other Artifact types.
To use MCP servers with Claude or ChatGPT, you need a paid plan with them (Claude Pro or higher, ChatGPT Plus or higher).
FAQ
-
Yes, on the Enterprise plan, with permission sets (Beta). For example, researchers can work with raw data while stakeholders only read published Artifacts. On other plans, each MCP client has one set of access permissions for everyone it's available for.
-
Yes. In the MCP client's Available for field, choose which users, roles or user groups can use it. You also decide what it can access, from published Artifacts only to raw data.
-
Only if the MCP client has the Read Participant data permission (Beta). Without it, participant data doesn't appear in any MCP response. Speaker names follow each Artifact's publishing settings, and redactions applied to a transcript carry through. Session information fields, like an email field, are part of the Session data the MCP can read with Read Projects and Sessions.
-
Not yet. The MCP can only create Notes for now.
-
Only if its MCP client has write permissions. Your AI tool is instructed to confirm with you before creating Highlights, and new Notes are always created unpublished, so you can review them before sharing.
-
No. MCP access, including raw data access, is included in your plan. Business plans can connect Claude and ChatGPT. Enterprise plans can also connect custom MCP clients and use permission sets.
