The official MCP server opens your transcripts to your assistant
{ "mcpServers": { "specala-ai": { "command": "npx", "args": ["-y", "specala-mcp-server"], "env": { "MEMO_API_KEY": "sk_live_…" } } } }
What tasks came out of yesterday's meeting?
From the May 28 meeting — 4 tasks: Anna — Friday release, Sergey — UI fixes, Maria — landing copy, Alex — ad budget.
Who owns the CRM integration?
Discussed on two calls: owner — Sergey, due June 10. Details in the May 22 recording, 11:40.
One config, any client
Claude
ChatGPT
Cursor
Visual Studio Code
JetBrains
Windsurf“What did we decide on pricing last week?” Your assistant finds the conversation and answers with a link to the exact moment — no re-listening
You
What did we decide on pricing last week?
On the Acme call (May 20) you agreed: base plan $29, enterprise on request. Alex locked the decision at 14:32.
“Did the reps cover every onboarding step?” Run calls against your checklist and get a summary of what was done and what was missed
You
Did the reps cover every onboarding step on these calls?
Checked 8 calls. Intro — 8/8, product demo — 7/8, pricing — 5/8. The “next contact” step was skipped in 3 conversations.
“Review my pitch: structure, clarity, filler words.” Analysis against your own criteria, based on the full timestamped transcript
You
Review my pitch: structure, clarity, filler words.
Structure is clear, but the intro runs long (2:40). Filler words: “like” ×12, “you know” ×9. Strong close with a call to action at 18:05.
“Which calls had complaints about onboarding?” Search decisions and details across hundreds of conversations at once
You
Which calls had clients complaining about onboarding?
Found 4: Acme (May 12), Globex (May 18), Initech (May 24), Umbrella (May 29). Shared pain — an unclear first step.
“Save this week's meetings to ~/notes in Markdown.” Clean Markdown with YAML frontmatter drops straight into Obsidian — up to 100 meetings per request
You
Save this week's meetings to ~/notes/meetings in markdown.
Saved 7 .md files with frontmatter. Ready to open in Obsidian.
Turn routine into a single phrase: “summarize this week's client calls and save them to the shared folder.” The assistant chains the tools itself
You
Summarize all client calls this week and save them to ~/team/weekly.
Processed 12 calls, summarized each, saved 12 files to ~/team/weekly. Ready to share with the team.
Export meetings to Markdown — they drop straight into your Obsidian. Local and within your assistant's reach
One key, and your assistant is ready to work with your meetings
In the web app, open Settings → API & MCP and click “Create key”. It starts with mk_live_ and is shown only once
Paste the ready-made config into Claude, ChatGPT or Cursor. For Claude Code it's a single terminal command — no file editing
“Find my meetings this week and collect action items.” Your assistant reaches out to Specala AI and answers from your real transcripts
Tools the assistant calls on its own — you just describe the task
Search by title, summary, speakers and topics — with filters for date, project and language
Complete text with speakers, timestamps and ready AI reports for a single conversation
Exports a meeting as .md or .txt straight to your folder — keeping the chat light and saving tokens
Up to 100 transcripts in a single request — export an entire project at once
Lists workspace projects so you can filter meetings by the exact track you need
The assistant relies on the real timestamped transcript, not a paraphrase — jump straight to the right second of the conversation
Search runs across your whole archive at once — decisions and commitments surface from dozens of conversations in one request
Long meetings are saved to disk as a file rather than dumped into the chat — the dialog stays light and tokens don't burn
The same mk_live_ works in Claude, ChatGPT and Cursor. No server to run, no integration to write — paste the config and go
The short version on setup, access and security
Create a key, paste the config — and ask your assistant about any conversation within minutes
