otter-axi

An agent-facing CLI (AXI tool) that wraps Otter.ai's hosted MCP server to find and pull meeting transcripts from any shell — headlessly, after a one-time browser login. Token-efficient TOON output, built for agents.

It does find-and-pull only; analysis is left to whatever consumes the output.

Why

Otter's hosted MCP connector re-prompts for auth and can't be scripted, and its Public REST API is gated to Enterprise workspaces. otter-axi turns the same sanctioned MCP server into a scriptable CLI: approve once in the browser, then search/fetch run non-interactively from any agent session or shell, with refresh handled silently.

Install & authenticate

npm install -g otter-axi       # install the CLI on your PATH
otter-axi auth login           # one-time browser approval
otter-axi doctor               # config → credentials → MCP reachable

Or run any command ad hoc without installing: npx -y otter-axi <command>.

Tokens are stored in ~/.config/otter-axi/config.json (mode 0600) and refreshed automatically; they are never printed. Read-only scopes: profile:read, conversations:read.

Usage

Find or browse meetings (returns metadata: title, date, duration, summary, action-item count, id). Empty query + a date range is browse mode:

otter-axi search "pricing discussion" --after 30d
otter-axi search --after 2026/05/01 --before 2026/05/07
otter-axi search --in-transcript "roadmap,milestones" --attended-by "Jane Doe" --mine

Flags: -q/--query, --after/--before (ISO or relative 7d/2w/3m/1y), --title-contains, --in-transcript, --attended-by, --channel, --folder, --mine, --limit, --full.

Pull a transcript by id (from search) or otter.ai/u/<id> URL:

otter-axi fetch <id>                 # metadata + preview
otter-axi fetch <id> --full          # verbatim transcript to stdout (for piping)
otter-axi fetch <id> --json-out      # parsed segments → auto-path in the OS temp dir
otter-axi fetch <id> --json-out=t.json  # …or an explicit path
otter-axi fetch <id> --csv-out          # CSV (--tsv-out / --text-out likewise)

The segment formats (--json-out/--csv-out/--tsv-out) parse the [H:MM:SS] Speaker N: … transcript into {start, speaker, text} rows — otter-axi owns one tested, lossless parser so agents don't re-derive it. Per the AXI side-channel-file convention, the path is optional (bare → <os-tmpdir>/otter-axi/<ts>-<id>.<ext> — transient scratch the OS cleans up), and writing a file is additive: stdout keeps the preview and adds wrote:/columns: and a jq hint.

Commands

| Command | Purpose | |---|---| | otter-axi | Content-first home / session status | | auth login [--no-wait\|--wait] | One-time OAuth; --no-wait/--wait for agent relay | | auth status [--offline] | Show the connected account | | auth logout | Revoke + clear stored tokens | | search [query…] [flags] | Find/browse meetings (metadata only) | | fetch <id\|url> [--full\|--text-out\|--json-out\|--csv-out\|--tsv-out] | Pull a transcript (verbatim or parsed segments) | | doctor | Tiered connectivity diagnostics | | setup hooks | Install the SessionStart hook (Claude Code / Codex / OpenCode) |

Development

Built on axi-sdk-js + @toon-format/toon + @modelcontextprotocol/sdk, TypeScript/ESM, bun for dev, tsc→node to ship.

bun install
bun run dev <command>   # run from source
bun run test            # vitest
bun run build           # tsc → dist/

Spec-driven: specs/ is the source of truth, plans/ tracks the work that built it. See specs/README.md.

License

MIT