Read a stored meeting, open its transcript and export the parts you need. This guide helps you choose the right command and recognize whether it reads local data, contacts Zoom or changes something. Install and connect first if no meetings have been imported or pulled.

zm meetings list --json
zm meetings show <meeting-id> --json
zm meetings transcript <meeting-id> --json
zm meetings summary <meeting-id> --json

Replace <meeting-id> with an ID from the list. Check the date and title before relying on its text. A missing summary or transcript means that part was not supplied, not that nothing was discussed. Stored reads use your selected local profile and do not fetch newer Zoom data. Use the archive guide to retrieve missing periods or import files.

To save a portable copy of the available content:

zm meetings export <meeting-id> --format markdown --json

Read the returned export before saving it to a file. MCP can return exports as text but does not write files. Search and evidence keeps exact passage references when you need citations.

You have an upcoming Zoom call and want the relevant earlier discussion beside it. Ask your agent:

Prepare for the next Research Example planning call in Zoom. Identify the intended series first.
Show the upcoming time, cited passages from earlier calls, confirmed project links, saved decisions,
notes about linked people/projects and directly sourced follow-up tasks. Show candidates if several
series fit. Tell me what is missing or uncertain. Do not change anything.

Use the provider series ID from the Zoom schedule, with digits and no spaces, rather than the internal ID printed by meetings list. In a terminal:

zm meetings prepare 123456789 --json

This reads the selected series from Zoom and joins it to your selected local profile's stored history. It needs the optional meeting:read:meeting permission in your Zoom app. It does not pull missing transcripts, change app permissions or start login. The default window looks seven days ahead and thirty days back, including at most five prior occurrences, five related meetings and ten cited cues per meeting.

Check item.upcoming.candidate and verified_next: a candidate time can be useful without proving it is the next occurrence. Recurring series dates alone never establish that claim. The result's meetings hold meeting references, transcript cues and fingerprints, observed participant names with existing person links, confirmed projects, accepted meeting facts and directly sourced tasks. Past participants are not a prediction of who will attend next.

related.items adds stored meetings outside this series through existing person links or confirmed project memberships shared with included history. Each has a meeting reference, cited cues and shared_person/shared_project flags; related.anchors shows the prior calls used to establish those connections. The search stays within this profile and history window, uses current links, and excludes proposed project memberships. Matching names or transcript words never creates a link.

knowledge.items adds accepted facts about explicitly linked people and confirmed global projects, including decisions recorded outside these meetings. Each has a fact: reference, subject/project references, update time and shared_person/shared_project flags. It uses the included prior calls as anchors and excludes proposed, reversed, deleted and later-updated facts. The fact reference lets you retrieve the full record and evidence through Memo before relying on it. This section contains accepted facts; knowledge.coverage keeps fact_evidence: not-revalidated and other_sources: not-queried visible.

notes.items adds owner-written notes and annotations about those linked people and confirmed global projects. Each includes a note: reference, title/text, revision, update time and a qualifying person/project reference. These are saved comments, separate from accepted facts. The command uses current revisions and confirmed associations; it excludes deleted notes, bot-authored notes, account-scoped subjects and revisions updated after the preparation start. Retrieve the referenced note in Memo when you need its full associations.

Check notes.coverage.has_more and notes.coverage.output_limited for additional or omitted notes. source_evidence: not-revalidated means the note's supporting sources have not been checked; other_sources: not-queried keeps imported documents, mail and messages outside this result's coverage. Selecting --json=notes.items.ref keeps that coverage visible beside the references.

If the upcoming time appears without earlier calls, pull the relevant period through the Zoom API so the store has its exact series links. Imported transcript files may not contain that provider series ID. Preparation does not match calls by similar titles or participant names.

Coverage distinguishes missing stored history, undated occurrences, extra rows and output limits. Archive completeness remains unknown. Imported documents, mail, messages and project-name/alias candidates remain unqueried; without usable prior occurrences, related history also remains unqueried. Stored fact evidence is not revalidated. Use search and evidence to investigate those gaps. Preparation does not generate a summary, record a decision, accept a finding or create a task.

For more control, use ISO timestamps with a zone for --since-time, --before-time and --history-since-time. --meetings caps each of exact-series and related history separately; --cues and --context-limit bound the context; --max-read-bytes bounds individual reads and --bytes bounds the entire preparation item. Exact-series evidence takes priority; related.has_more and related.output_limited identify extra or omitted related context. Accepted knowledge has independent knowledge.coverage.has_more and knowledge.coverage.output_limited flags; it returns at most twenty facts (or a smaller --context-limit) and yields bytes to meeting evidence. Notes have their own lookahead and output-limit flags, with at most twenty entries (or a smaller --context-limit). They fit after accepted facts and meeting evidence. Additional rows may remain when a bound is reached. Skipped unavailable sources set meta.complete=false; inspect their references before relying on the brief. Increase the relevant bound when the coverage identifies truncation.

A successful local read does not prove Zoom still has the same content. A successful cloud download does not import that transcript into the meeting database until you request archive import.

Stored meetings commands use local database ids; remote-meetings uses Zoom's numeric ids. Managing Zoom meetings covers the write commands.

Use --json for one complete answer. A single result has item; a list has items and page (mode, limit, has_more, optional next_cursor). Both carry meta.run_id, meta.duration_ms and meta.complete, with optional next steps and missing information. Result keys use snake_case.

Detailed reports, including participant discovery, evidence, semantic search and unified search, keep their coverage and report-specific continuation inside item. Stored meeting word search returns standard hit items with ref, meeting, excerpt and offset_ms. When stored, source adds the meeting title, started_at, evidence ref and document url. A null URL means no source permalink is stored; a meeting join link is not an evidence link. Meeting details use item; transcripts and summaries use items. Details count current transcripts; use meetings transcript <meeting> --history to read retained older revisions.

--jsonl emits typed records: each result is {"type":"item","data":...}, followed by an {"type":"end","complete":true,"page":...} record. Empty lists still emit the end record. Reports stay together inside one item record; each streaming acknowledgement has its own end record. Read complete and optional issues on the end record. Use --json when you need the full answer's next, missing and execution metadata. Command discovery is the shared contract-3 document with separate effects.remote and effects.local declarations; help, version and shell completion retain their protocols.

A skipped source makes meta.complete false and adds a source issue. search all accepts partial results and exits zero; inspect its skips before treating the result as complete. Other partial answers exit with the first issue's error code after printing the available result. A bounded page with more results is ordinary pagination, rather than a source failure. Errors go to stderr. Use --quiet to suppress non-error diagnostics. Piped output defaults to JSON; terminal output uses the shared human-readable renderer. Completion emits its shell protocol unless JSON is requested.

Use --json=id,title to keep only those fields. Dotted paths select nested objects and arrays, for example --json=ref,excerpt,source.title,source.started_at,source.url for meeting search hits. Paths are relative to the answer payload; pagination, coverage, next steps, missing data and execution metadata remain. Projection reduces printed output, not the work scanned. --offline refuses network requests. Finite commands buffer successful output until resource cleanup completes, with a default output limit of 4 MiB; --max-output-bytes 0 removes that limit.

--timeout <milliseconds> sets a positive whole-command deadline. Durations also accept units such as 30s, 5m and 1h. Defaults are 30 seconds for reads, five minutes for login and one hour for recording downloads, sync and archive verification. Foreground watching and MCP serving have no default deadline. Cancellation waits for active operations and cleanup; an injected operation that ignores its signal can delay shutdown.