Skip to content
github-actions[bot] edited this page Nov 16, 2026 · 1 revision

MCP and skills

Let an AI agent read your stored meeting text and return source-backed answers. Choose a skill for a terminal agent or MCP for an AI app that launches local tools. A skill is an instruction file; MCP is the protocol the app uses to call tools. Neither connection logs in or downloads meetings. Import or pull at least one meeting before trying a transcript question.

Choose the connection

Agent you use Connect it Result
Codex or Gemini CLI on this computer zm skill install --for agents Installs instructions for using CLI commands
Claude Code on this computer zm skill install --for claude Installs its CLI instructions
An MCP-capable local AI app Add the server entry below and restart the app The same commands, as three tools

Both reach the same commands under the same permissions of the profile.

The agent skill

zm skill show prints the packaged instructions for an agent. zm skill install writes a version-stamped copy to the agent skills directories: ~/.claude/skills/zoom-cli/ for Claude Code and ~/.agents/skills/zoom-cli/ for Codex and Gemini CLI. --for claude or --for agents selects one; the default is both. Installation is explicit and does not run when a command starts.

The MCP server

zm personal mcp serve serves zm's commands over MCP's stdio protocol as three tools, the same three every WireCat CLI serves:

Tool What it does
zm_tools_search finds the command for a task — "meetings list", "search", "prepare" — with its arguments
zm_read runs a command that changes nothing: { "command": "meetings list", "arguments": {} }
zm_write runs a command that changes something, such as meetings create

A call runs the command line itself, so it answers and fails exactly as zm personal <command> --json does. The server speaks for the profile named in its args; a caller cannot change it. watch and the server itself are not offered. Shared inventory (meetings scheduled), project scanning and upgrade are available under the profile permissions. Interactive login (session start) stays in the terminal. Settings, skill installation, upgrade and archive timer changes require mcp.authority at allow or ask; profile permissions can only be changed in a terminal. Grant authority with zm personal config set permissions '{"mcp.authority":"allow"}' and reconnect the app if needed. This replaces the profile's permission map, so include any existing restrictions.

Denied commands remain in full listings and searches with availability.available: false, a reason and the permission key. readonly marks writes unavailable. Named-profile CLI help and discovery show the same current permission information. Calls recheck permissions before running. A command at level ask runs without a question, since no one is at a terminal: the built-in ask on meetings create, meetings update and meetings cancel lets an agent change Zoom meetings. To keep an agent from doing that, give the profile it serves {"meetings.create": "readonly", "meetings.update": "readonly", "meetings.cancel": "readonly"}.

In the mcpServers format of Claude Desktop's configuration:

{
  "mcpServers": {
    "zoom": { "command": "zm", "args": ["personal", "mcp", "serve"] }
  }
}

A desktop app may not see your terminal's PATH, and a stdio server inherits only part of the environment (MCP debugging guide). If the server does not start, use the full path from command -v zm on Linux/macOS or where.exe zm on Windows as the command. If you set ZM_CONFIG_DIR or MESSAGING_STORE, pass the same values in the entry's env, or the server reads a different profile or database. Keep stdout for protocol messages; omit --json and --jsonl.

Make one server stricter

zm personal mcp serve --permission meetings=deny makes a permission stricter for that server only, and repeats for more keys. It never loosens the profile: a value that would is refused when the server starts, and so is a key that names no command. zm personal mcp config and zm personal mcp add take the same flag, and mcp add and mcp check set up and test the entry for Codex or Claude Code.

Browser apps over HTTP

zm personal mcp serve --http --public-url https://<name>.ts.net serves on 127.0.0.1 for AI apps in a browser, behind your own tunnel; --port changes the local port (default 8765). Writes follow the profile's permissions, as over stdio. A browser app logs in once with a code that the server prints on stderr (mcp server status shows it again); its access lasts one hour and its refresh 30 days. zm personal mcp revoke ends every login. zm personal mcp server start, stop and status run that server in the background, and --profiles works with all of these.

Several profiles from one server

zm mcp serve --profiles work,personal serves the profiles it names from one server, and zm mcp serve --profiles all serves every profile here, so an AI app needs one Zoom entry, not one per profile. Each tool then takes a required profile, and the agent is told to ask when you named none. A call runs in that profile's own zm <profile> mcp serve, with that profile's permissions. The server names the profiles it serves on stderr when it starts.

A server always names what it serves: zm mcp serve with no profile and no --profiles stops and lists the profiles here, and so do a profile together with --profiles and a name that is not a profile here. Under ZM_PROFILE_LOCK, --profiles may name only the locked profile.

{
  "mcpServers": {
    "zoom": { "command": "zm", "args": ["mcp", "serve", "--profiles", "all"] }
  }
}

Try the connection

After installing the skill, start a new agent session if it has not discovered the instructions. After adding MCP, restart the AI app and confirm that its Zoom tools are available. Ask:

List my stored meetings. Open the latest one and show its date, available transcript and one
passage supporting its main decision. Tell me if the data is incomplete.

Check the selected meeting against zm personal meetings list --json in your terminal. If discovery works but no data appears, verify the same profile and store path, then pull or import data. A remote web chat cannot run this local command unless it has a supported connection to your machine.

Selected communications in meeting preparation

meetings prepare reads the message, mail and document sources configured for the profile (source selections); over MCP it is the same command, so the same sources apply and a source not configured is refused. See limits and continuation.

Clone this wiki locally