Picture a Monday at 9:00, management committee of a 40-person SME. They decide to raise prices by 6% in January, Laura commits to sending the proposal to the big client on Thursday and Pedro to reviewing the supplier's contract. On Thursday nobody remembers whether the 6% was for everyone or only for new clients. The summary is in the account of a notes tool belonging to whoever turned it on, who went on holiday. And the company's AI, the one that answers the team, knows nothing about that meeting.
What is valuable stays outside
A company can have a very good AI and still be only as smart as what it knows. The most valuable things that happen in the company (client meetings, committees, interviews) almost never reach that AI: they live in recordings, in personal notes or on a vendor's platform.
- The summary belongs to an employee, not to the company: if they leave, they take it.
- Each notes tool is a closed box with its own access: for example, Otter lists the API and webhooks only on its Enterprise plan, and Granola offers its API from the Business plan, according to their pricing pages.
- Copying and pasting summaries by hand does not scale.
What it takes to solve it
- Each meeting should be structured data: a transcript with the real speaker, decisions, tasks and commitments.
- That data should sit in the organisation and be queryable by API or MCP from the AI you already use.
- It should feed a company memory, with personal data redacted.
How each option on the market solves it
- Otter: has an MCP server; the API and webhooks only on Enterprise.
- Fathom: public API and MCP, with integrations toward CRM and sales tools.
- Read.ai: MCP, and API and webhooks on premium plans.
- Granola: API from Business and MCP with limits depending on the plan.
- VideoSuite: API v1 and MCP from the Professional plan, and delivery of each report to MemoryFirst; no outgoing webhooks for now.