WireCat Zoom

[Zoom meetings and transcripts for your AI agent]

zoom-cli

Zoom meetings and transcripts for your AI agent. Know the context before the call. Remember the commitments after it.

zm keeps your Zoom meetings, transcripts and AI summaries in a local database and gives them to Claude Code, Codex, Gemini CLI or any MCP client. Ask what you discussed with someone last time, or what you promised in yesterday's call, and get the answer with a reference to the exact passage.

Documentation · Commands · WireCat

[npm] [Node] [License: Apache 2.0]

Example — invented people, illustrative output:

You: I have a call with Bob Sample at three. What did we agree last time, and what do I owe him?

zm personal remote-meetings list --json                      # your upcoming meetingszm personal meetings people "Bob Sample" --json              # him among past participantszm personal search meetings "quote" --since-time 2026-09-01 --jsonzm personal meetings passage meeting:1/42/57/118 --format markdown  # the line and its context

Agent: You last spoke on 2 October in "Pricing review". You promised Bob a revised quote by Friday (meeting:1/42/57/118), and he asked whether the annual plan includes onboarding (meeting:1/42/57/131). I found no later update in the stored meetings I searched; missing history may change that.

Quick start

You need Node 22.16+ (22.x) or 24+ and a configured user-managed Zoom app; a system keyring is best. Reuse one you already created or one your team has made available to your user; otherwise create your own. Use its Public Client ID, not its regular Client ID or secret. The installation guide walks through app settings, browser approval and callback completion.

npm install -g @wirecat/zoom-clizm config set zoom.client_id <public-client-id>zm personal session start                 # "personal" is the profile; approve in your browserzm personal meetings sync --lookback-days 7zm personal meetings list

Check Zoom's transcript generation and retention settings first: the API reads available hosted meeting text, and permission does not create a transcript. For an attended meeting, obtain a VTT file from the host or download one if allowed, then run zm personal meetings import <folder>. Full guide: installation.

More: installation · your Zoom app · Zoom on wirecat.dev

What you can do

Prepare for a call. "I meet Bob Sample tomorrow. What did we talk about last time?" meetings scheduled finds candidate series; meetings prepare <series-id> checks the selected series’ next occurrence and gathers cited context. meetings people, search meetings (word search that also matches English, Spanish and Russian word forms) and meetings context person:<id> bring back past meetings with the same people. Inventory date hints alone do not prove the next call. Participant discovery accepts --limit, --scan-limit, --max-read-bytes and --cursor; follow item.page.next_cursor even after an empty scan page. It returns observed identity_id candidates, never inferred person links. Stored meeting lists accept metadata queries, for example zm personal meetings list 'started_at:[2026-09-01 TO 2026-10-01]' --json.

Find the project behind a call. "Which of my saved projects might this meeting concern?" meetings projects scan <ref> compares one bounded page of local meeting evidence with existing project names, keys and aliases. It returns cited suggestions and ambiguity for review, saving no project link. The agent can check proposed labels against Memo's existing vocabulary too.

Follow up after it. "What did I promise in today's call?" The agent reads meetings transcript and meetings summary, cites passages with meetings evidence, and meetings task-proposal --type promise previews a follow-up task tied to that passage.

Keep the whole history. "Back up every recording since January." recordings download --since-time 2026-01-01 --until-time 2026-10-10 --output-dir ./zoom-archive --jsonl downloads video, audio, transcripts and chat with SHA-256 manifests; recordings schedule runs it daily.

More: the local archive · search and evidence · Zoom on wirecat.dev

Built for AI agents

• Command line + skill. zm skill install adds the packaged guide for Claude Code, Codex and Gemini CLI. Every command speaks --json or --jsonl, and zm commands --json describes them all.
• MCP. zm personal mcp serve serves zm's commands to Claude Desktop, Cursor and other MCP clients, as far as the profile's permissions allow; zm mcp serve --profiles all serves every profile from one server, and --http serves browser apps behind your tunnel. Connecting a client.

Your agent. Your meetings. Your rules.

More: MCP and skills · connect your AI agent

Why zm

• Context across meetings. Search available hosted meetings and imported transcripts in the shared store. Read archived transcripts separately or import them before searching them together.
• Local-first. Meetings live in a SQLite database on your machine. Archived transcripts can be read and searched offline, without login.
• Every answer has a source. References such as meeting:1/42/57/118 point to one passage of one transcript revision, and corrected transcripts keep their history.
• Made for agents. JSON output, bounded reads with explicit completeness, a skill and an MCP server.
• Safe by default. Your own Zoom app, tokens only in the keyring, permissions per profile, and every change to Zoom asks first at the terminal. Over MCP there is no terminal, so ask runs without a question; use deny or readonly to stop a change.

More: security · Zoom on wirecat.dev

What works today, and what does not yet

zm is at version 0.x; commands may still change. Today it reads hosted meetings, participants, transcripts and AI summaries available to the account through Zoom's official API, imports downloaded WebVTT files, archives cloud recordings, transcribes local audio with an installed speech model, links meetings to events and people, and creates, changes or cancels simple meetings.

• It does not find commitments on its own. Your agent finds them in transcripts; task-proposal only previews a task and creates nothing.
• People are linked by hand. zm never guesses a person from a speaker name; person context covers participants already linked to a person, for example with meetings link.
• Local transcription needs ffmpeg for anything but a ready WAV, and a model you download with zm models download; speakers stay unknown.
• Attended meetings need a manual download from your Zoom portal, then zm personal meetings import.
• A company-managed Zoom account may need an admin to approve your app.
• Daily archive jobs install only on Linux. Other platforms can preview them.
• Zoom only, for now. Google Meet and Microsoft Teams are on the roadmap.

More: what is planned · Zoom on wirecat.dev

More: compared with other tools · on wirecat.dev

Commands by task

Zoom meeting changes, links, transcription, schedules, explicit meetings embeddings generation and lock recovery preview by default and run once with --yes; --dry-run keeps the preview. Word search updates its index automatically. Search with --model generates missing vectors before ranking; an explicitly selected remote embedding endpoint receives those transcript passages and the query.

More: usage · every command and option · Zoom on wirecat.dev

Your Zoom app and login

zm reads your meetings through Zoom's official API using a configured user-managed Zoom app with Public Client OAuth. Reuse an existing app available to your user or create one. Enable Public Client OAuth, copy its Public Client ID, configure the redirect and add the required scopes manually. The optional starter manifest supplies neither scopes nor Public Client OAuth. Zoom may show a Client Secret, but zm does not use it. Recording archives and management need additional scopes. Follow the app setup guide; no credentials file is downloaded.

zm personal session start opens Zoom's consent page. Zoom redirects to the static page https://wirecat.dev/zoom-callback, which passes the answer to a listener on your machine; if it cannot, paste the code=…&state=… parameters it shows. Login checks the state and PKCE verifier and stores tokens in the system keyring; where the keyring fails, in a private file, without asking (--credential-storage file chooses the file, not recommended). Without a terminal, as in an agent's shell, login prints the URL on stderr for the agent to pass on and waits for the callback; --no-input and CI refuse it. zm personal session status reads the cached state without refreshing it; --check asks Zoom.

More: your Zoom app · login and sessions · Zoom on wirecat.dev

Profiles and the local store

The first word is the profile: zm personal meetings list. A profile is one Zoom login and its own meetings in the store; zm work session start makes a second one. ZM_PROFILE names it for a shell session, and ZM_PROFILE_LOCK pins a process to one profile. Nothing picks a profile for you: without one, a command that uses an account stops and lists the profiles on this computer. Names have 1–64 letters, digits, dots, underscores or hyphens. Login makes a profile; so does the first import on a computer with none. An import or pull naming an unknown profile beside others is refused, so a typo stores nothing.

Meetings live in the shared WireCat database wirecat.db, next to data from other WireCat tools. MESSAGING_STORE selects another path. Store opens preserve legacy messages.db files and their sidecars; manage their retention separately. Metadata, help, configuration and login never open the database.

More: profiles · where meetings are stored · Zoom on wirecat.dev

Pull, import and watch

zm personal meetings sync --lookback-days 7                       # hosted meetings, revisiting late transcriptszm personal meetings import ./downloaded-transcripts              # .vtt or zoom-searcher .txt fileszm personal watch api --lookback-days 7 --jsonl                   # keep pulling in the foreground

The cursor alone does not revisit transcripts that appear late; use --lookback-days or an older --since-time. A failed occurrence leaves the cursor unchanged for the next pull. Imported files follow the exporter's YYYY-MM-DD_HHMM_topic-slug.vtt names, or .txt with a zoom-searcher header; their time is read in this computer's zone unless files.timezone names another, and the import says which. A .txt's speakers become participants of that meeting only; speaker names never become identities across meetings.

More: the local archive · Zoom on wirecat.dev

Archive cloud recordings

zm personal recordings download --since-time 2026-01-01 --until-time 2026-10-10 --output-dir ./zoom-archive --jsonlzm personal recordings archive check ./zoom-archive --offline --jsonzm personal search recordings ./zoom-archive "example plan" --offline --json

Sync covers up to ten years in monthly requests and downloads serially, streaming to disk. Each occurrence gets a manifest with sizes and SHA-256 checksums, never signed URLs or tokens. Existing files are never overwritten; repeated runs skip verified files. With a profile, interrupted downloads resume safely. recordings archive transcript and search recordings read verified WebVTT offline without a database; recordings archive import appends them to your stored meetings. recordings schedule create --yes installs a daily systemd user timer (Linux only).

What you can download depends on Zoom permissions, retention and your plan. Recordings of meetings you only attended are not necessarily available through your API.

More: archive cloud recordings · daily backups · Zoom on wirecat.dev

Search and evidence

zm personal search meetings "example plan" --limit 20 --json       # word search with continuationzm personal meetings evidence meeting:3/15/20/0 --cues 100 --json  # the exact passagezm personal meetings context person:7 --meetings 10 --json         # meetings linked to a person

search meetings indexes new meetings first, matches every word by prefix or by stored English, Spanish and Russian word forms, and pages with an opaque --cursor. References have the form meeting:<account>/<meeting>[/<transcript>[/<cue-position>]], stay inside the selected account and keep pointing at the same revision after a correction. Bounded reads report their coverage and limits; archive completeness can remain unknown. search all --only meetings,notes adds store-wide notes. search meetings --model searches by meaning with a model you name; nothing is downloaded for you.

More: search and evidence · topic search · Zoom on wirecat.dev

Local transcription

No Zoom transcript? zm personal meetings transcribe 1 ./call.m4a --model <model> --yes recognizes a recording — Zoom's MP4 or M4A through an installed ffmpeg, a 16 or 48 kHz PCM16 WAV as it is — with a local speech model from zm models download <model>, and appends it as a separate transcript. The preview checks the file without loading the model. Timestamps cover windows of up to 30 seconds and speakers stay unknown.

More: transcribe local audio · Zoom on wirecat.dev

MCP and agent setup

zm skill install writes the packaged guide to ~/.claude/skills/zoom-cli/ (Claude Code) and ~/.agents/skills/zoom-cli/ (Codex, Gemini CLI); --for picks one. MCP clients start the server with the profile it speaks for:

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

The server offers zm's commands as the three tools every WireCat CLI serves — zm_tools_search, zm_read, zm_write — and runs each call as the command line would. The profile's permissions decide what it offers: {"remote-meetings": "readonly"} keeps an agent from changing Zoom meetings. A desktop app may need the full path from command -v zm (Linux/macOS) or where.exe zm (Windows), and the same ZM_CONFIG_DIR or MESSAGING_STORE in the entry's env.

More: MCP and skills · MCP and skills on wirecat.dev · connect your AI agent

Manage Zoom meetings

zm personal remote-meetings create --title "Example planning" --start-time 2026-10-20T10:00:00Z --duration-minutes 30 --dry-run --jsonzm personal remote-meetings cancel 123456789 --dry-run --json

Creation, updates and cancellation preview until you add --yes, and are never retried automatically. An uncertain outcome exits with code 14: check Zoom before repeating. Remote ids are Zoom's numeric ids; stored meetings commands use local ids.

More: managing Zoom meetings · Zoom on wirecat.dev

Output, limits and deadlines

--json prints one answer envelope; --jsonl prints typed item records and a final end record; piped output defaults to JSON and errors go to stderr. --json=id,title keeps only those fields, --offline refuses the network, and output is capped at 4 MiB unless --max-output-bytes says otherwise. Default deadlines are 30 seconds for reads, five minutes for login and one hour for downloads, sync and verification; --timeout 5m changes them. Watching and MCP serving run until stopped. Times are UTC in every format; text output that shows one ends with times in UTC.

More: usage · Zoom on wirecat.dev

Configuration

More: configuration · Zoom on wirecat.dev

Security model

• A configured Zoom public-client app with PKCE; no client secret is used.
• Tokens in the system keyring, never in a config file; where the keyring fails, an owner-only file, said once on stderr.
• Network: Zoom's API, plus an embedding endpoint only when you name one. --offline refuses both.
• Manifests and saved jobs never hold signed URLs, tokens or credentials; diagnostics omit raw provider error text.
• The MCP server offers reads and writes for the profile it serves (or each profile --profiles names), limited by its permissions; ask writes run without a terminal question.
• ZM_READ_ONLY=1 refuses changes to Zoom, the login, settings and anything you added (imports, links, transcripts, exports, jobs), with exit 5, before any network call; reads and the syncs that copy Zoom into the store or an archive still run. Give it to an agent that should only read.

More: security · security on wirecat.dev

Troubleshooting

• "Invalid redirect" from Zoom: the manifest import left the redirect URL or Public Client OAuth unset; fill them in by hand.
• No transcripts: check generation, retention and permissions in the app guide. Enabling a setting does not restore deleted text; attended meetings may need zm personal meetings import.
• Missing-scope error: add the named scope in your app, then zm personal session start again.
• Browser does not start: zm personal session start --browser firefox, or paste the callback parameters.
• Lock left after a crash: zm personal recordings archive locks, then recover with --yes.

More: troubleshooting · Zoom on wirecat.dev

Part of WireCat

zm is part of WireCat, an open-source, local-first context layer for AI agents. Meetings land in the same local store as your Telegram, MAX and memo data, and zm personal search all "<query>" --only meetings,notes searches them together with your memo notes. Start with Zoom. Add more context when you need it. Your context. Any agent.

More: wirecat.dev · connect your AI agent

Questions and setup help: join the WireCat Discord. Bugs go to this repository's issues.

Technical docs

The Zoom guides on wirecat.dev include setup, search and archive instructions.

Development

Node 22.16 or newer within 22.x, or Node 24 or newer:

pnpm installpnpm lint && pnpm typecheck && pnpm test:coverage && pnpm docs:checkpnpm build && pnpm smoke:bun

Fixtures use invented people and meetings. Tests use an injected keyring, scripted HTTP responses, temporary config folders and the memory meeting store. pnpm generate:check checks the generated command reference; pnpm test:matrix measures command and option coverage.

More: development setup · test matrix

Licence

Apache License 2.0.