Skip to content

The MCP server and connector for your meetings.

Connect Kalima to Claude, ChatGPT, Cursor, or VS Code. Your assistant can search your transcripts, find what was said in a call months ago, and point you to the meeting it came from. Setup takes about two minutes, uses OAuth, and gives the assistant read-only access.

Server URL
https://mcp.meetkalima.com/mcp
Connect your AI toolSee the 12 tools
Kalima MCP

What did we agree on about the Q3 pricing change, and who raised concerns?

search_utterances

Across three meetings, the team settled on a 12 percent increase for new customers only, with existing accounts held until renewal. Priya raised the main objection in the June 14 revenue sync, arguing that the enterprise tier needed its own timeline.

3 meetings referenced

Add the connector once, then just ask.

There is nothing to install and no API key to manage. Your AI tool connects directly to Kalima through the Model Context Protocol.

  1. 01

    Add the server URL

    Paste the Kalima MCP URL into your AI tool's connector settings. The supported tools on this page can all connect to a remote MCP server.

  2. 02

    Sign in and approve

    Your browser opens a Kalima approval screen. Approve read-only access once, and the connector uses a limited token from then on.

  3. 03

    Ask about your meetings

    Your assistant can now search transcripts, open a specific meeting, and show the exact passage it is quoting, all without leaving the chat.

Set the connector up in your tool of choice.

Step by step guides for the five AI tools people connect most. Whether your tool calls it a connector or an MCP server, the setup is the same and takes about two minutes.

Connect Kalima to ChatGPT

ChatGPT can add Kalima as a custom plugin once developer mode is on. The setup takes a minute, but the switch is not where most guides say it is, and one URL detail catches almost everyone out.

Requires
Developer mode enabled
Method
Custom plugin

  1. Turn on developer mode

    Open Settings and choose Security and sign-in from the left sidebar, not Connectors. Scroll down to the Developer mode section and switch it on. ChatGPT marks it as elevated risk, because developer mode is what allows unverified connectors to be added at all.

  2. Open Plugins and start a new one

    Back in the main ChatGPT window, click Plugins in the left sidebar. On the Plugins page, use the round plus button at the top right to open the New plugin dialog.

  3. Fill in the two fields at the top

    Name it Kalima. In Description, say briefly what it is for, for example: read-only access to my Kalima meeting transcripts. The description is optional, but it is what tells the model when to reach for this plugin, so it is worth writing.

  4. Paste the server URL, ending in /mcp

    Under Connection, leave Server URL selected and paste the endpoint below. Leave Authentication set to OAuth. Ignore the greyed-out example, which shows an /sse address: Kalima uses streamable HTTP.

    Server URL
    https://mcp.meetkalima.com/mcp

    Keep /mcp on the end. ChatGPT accepts a bare domain and will even complete the sign-in successfully, because it finds the metadata there, but the plugin then lists no available actions. If that happens, delete it and recreate it with the full URL.

  5. Confirm and create

    Tick the box confirming you want to continue, then press Create. ChatGPT sends you to Kalima to sign in and approve read-only access, and the Kalima tools become available in a chat.

Using something else?

Kalima implements the Model Context Protocol over streamable HTTP with OAuth 2.1, so any spec-compliant client can add it as a connector using the same server URL. Windsurf, Zed, Cline, and Goose all work the same way: add the remote server, sign in, approve.

Twelve tools, one connection.

Your assistant picks the right tool for the question. You never call these by hand.

Tool
Finding meetings
search_sessionsFull-text search across every meeting you can access, ranked, with a snippet and timestamps.
list_sessionsBrowse meetings by recency, length, or name, filtered by project, team, status, or date range.
search_utterancesFind the exact moments a topic came up across all your meetings, with speaker and timestamp.
Reading transcripts
get_transcriptPaginated transcript for one meeting. Ask for full detail, speaker labels only, or plain text.
search_transcriptJump to where a topic was discussed inside a single meeting instead of reading the whole thing.
Meeting detail
get_sessionOne meeting's metadata: status, mode, start and end, duration, participants, word count.
Projects
list_projectsYour project folders and their sub-projects, with session counts and last activity. Searchable by name.
get_projectOne project in full, including total recorded time, its place in the folder tree, and its most recent meetings.
People and patterns
list_speakersEveryone who has spoken across your meetings, with how often and how recently.
get_activity_statsTotals across your account: meetings, recorded hours, words transcribed, first and last session.
Workspace
list_organizationsThe organizations you belong to and your role in each.
list_teamsThe teams you belong to, with role and parent organization.

Three built-in prompts

Ready-made flows your tool can offer as one-click actions, so common questions do not need to be typed out.

weekly_review

Walk through everything recorded in the past week.

catch_up_on_project

Get up to speed on one project's recent meetings.

find_mentions

Track down every time a topic, name, or account came up.

The answer is usually in a meeting you already had.

The information is usually there. Finding it is the hard part. Nobody wants to search through four hours of recordings for one sentence, so decisions get revisited and commitments slip through the cracks. Your assistant can find the relevant moment for you.

  • What objections came up most often in demos this quarter?

  • Summarize where the Meridian project stands based on our last four calls.

  • Did we ever promise that customer a custom SLA? Show me exactly what was said.

  • Draft the follow-up email from this morning's kickoff, using what we actually committed to.

  • Who has been in the most meetings about the migration, and what did each of them own?

  • Find the moment in Tuesday's review where we talked about the pricing tiers.

Read-only access, technically enforced.

Giving an AI tool access to meeting data requires trust. Here is exactly what the Kalima connector can and cannot do.

  • It cannot change your data

    The MCP server uses a dedicated database role with permission to read data only. Read-only transactions are enforced at the connection level. There is no write access to turn off because the connector never receives it.

  • No API keys to create or paste

    Kalima uses OAuth 2.1 with PKCE. You approve access in your browser, and your tool receives a limited token that expires automatically. Nothing sensitive needs to be stored in a configuration file.

  • It only sees meetings you can access

    The connector follows your permissions in Kalima and narrows them further. Password-protected share links and open link-sharing grants are intentionally excluded.

  • Revoke access whenever you need to

    Every connector you have approved appears under Account, then Connected apps. When you revoke one, access ends on its next request rather than waiting for the token to expire.

  • Separate from live transcription

    The MCP server runs in its own process with its own connection pool and capacity limits. A busy assistant cannot slow down a recording that is already in progress.

  • Your AI summaries are not exposed

    The connector does not provide Kalima's generated summaries. Your assistant works from the actual transcript, not another model's interpretation of it.

Kalima Labs UG is based in Germany and holds SOC 2 Type II and ISO 27001 certification.

Included on every plan.

The Kalima connector works on every plan, the Free one included. How many requests you get per day and week depends on the plan.

PlanRequests per day
Free100
Plus250
Pro1,000
Team5,000
EnterpriseUnlimited

One request is one tool call, not one question. Answering a single question often takes your assistant two or three calls, so budget accordingly.

Questions, answered.

The Model Context Protocol is an open standard for connecting AI assistants to outside data. An MCP server exposes a set of tools that an assistant can call on your behalf. The Kalima MCP server exposes twelve read-only tools over your meetings, so your assistant can search and read transcripts without you copying anything into the chat.
In practice, yes. Connector is the word most AI tools show you in their own interface: Claude and ChatGPT both call these Connectors, while Cursor and VS Code call them MCP servers. They are two names for the same thing, and Kalima works as both. Add the Kalima connector, or add the Kalima MCP server, and you get the same twelve read-only tools either way.
Anything that speaks the Model Context Protocol. This page has step by step guides for Claude Desktop, Claude Code, Cursor, ChatGPT, and VS Code with GitHub Copilot. Windsurf, Zed, Cline, and Goose connect the same way using the same server URL.
No. Kalima uses OAuth 2.1 with PKCE, so you sign in through your browser and approve access on a consent screen. Your tool receives a scoped token that expires and rotates on its own. There is no key to generate, paste, or rotate manually.
No. Every tool is read-only, and that is enforced at the database level by a role that only has permission to read. The connection cannot write even if something above it tried to.
Exactly the meetings you can already open yourself: the ones you own, ones shared with you directly, and ones you can reach through a project, team, or organization. Password-protected links and open link-sharing grants are excluded, so the MCP view is slightly narrower than the app.
All of them, including Free. The Free plan gets 100 requests a day and 500 a week. Limits rise through Plus, Pro, and Team, and Enterprise is unlimited. If you hit a limit, the connector tells you which one and when it resets.
Yes. Each connector authorizes separately and appears as its own entry under Account, then Connected apps, so you can run Claude and Cursor against the same account and revoke either one independently.
Open Account, then Connected apps, and revoke it. Access is cut on that connector's next request, not whenever its token happens to expire.
Yes, as a custom plugin. Turn on Developer mode under Settings, then Security and sign-in, then add Kalima from Plugins in the left sidebar. One detail matters: enter the full endpoint URL ending in /mcp. A bare domain will connect and then show no available actions.
No, deliberately. The tools read transcripts, speakers, and meeting metadata. Your assistant works from what was said rather than from a summary another model already wrote.

Put your meetings where you already work.

Start recording with Kalima, add the connector to your AI tool, and stop rewatching calls to find one sentence.

Start for freeSee pricing