Revolut People has AI in four places: helping managers write reviews, summarising 1:1s, pulling evidence into reviews, and an AI Assistant that works with your People data. There is also an MCP server, currently in Beta, that lets AI tools such as Claude Code or Cursor work with Revolut People under your own permissions. This page sets out what each one does, with a link to the documentation behind every statement.
The AI features
AI help with review writing
When a line or functional manager writes a review, the platform can draft feedback from recorded 1:1s, with a 1:1 Summary panel alongside. Managers choose “Yes, use this” or “No, clear field”, and the docs say to always review AI suggestions before submitting.
Feedback Quality Indicator
As a manager types, feedback is rated Poor, Basic or Great against six checks: constructive, evidence-based, actionable, aligned with competencies, balanced in tone, and an appropriate length (around 50 to 100 words).
Performance Insights
Summaries of activity during the review cycle from goals and epics, Jira, GitHub, Bitbucket, 1:1s, manager survey scores and interviews. Only data inside the cycle dates counts, people are matched by email, and insights appear in line and functional manager reviews only.
Source: Performance Insights
1:1 summaries
1:1s held on Google Meet (with Gemini transcripts or the Meetings Extension) or Microsoft Teams (with transcription switched on) get AI-generated meeting summaries, which then support AI-assisted feedback during review cycles.
Sources: Gemini · Meetings Extension · Teams
AI Assistant
Since September 2026 the AI Assistant works with your People data as well as answering product questions: time off, pay and documents, personal details, people and performance (grades, scorecards, goals, KPIs), hiring, and turning a question into a saved report. It can also search Jira and Confluence through Atlassian’s Rovo MCP.
Sources: September 2026 product update · Reports with the AI Assistant
Permissions come first
The docs say the AI Assistant only sees what you can already see and only makes changes when you ask it to. Reports it builds use only data you have permission to access.
Sources: September 2026 product update · Reports with the AI Assistant
The MCP server Beta
MCP (Model Context Protocol) is an open standard that lets an AI tool call another product’s functions. The Revolut People MCP server is a Beta feature: the docs describe it as available under a limited release and coming soon to all customers (MCP setup).
- Clients with setup guides: Claude Code, Codex App, Cursor and OpenCode.
- Tools: 55 tools across compensation, documents, engagement, meetings, objectives, organisation, performance, personal info, recruitment and time off. Objectives has the most, with 13 (supported tools).
- Examples: read a goal with its KPIs and progress, create a draft goal and submit it for approval, list a manager’s direct reports’ performance for a cycle, generate a feedback draft for a direct report, or get notes and action items from meetings (objectives · performance · meetings).
- Permissions: you sign in and approve access through an OAuth consent screen. The server works within the same permissions as your Revolut People account and grants nothing beyond them. If your role changes, its access changes too, and an action you are not allowed to do returns an error (permissions).
Because it is Beta, check availability for your workspace on a discovery call before you plan around it.
The REST API
For your own integrations there is also a REST API that covers employees, departments, teams, roles, goals, scorecards, final grades, probation, recruitment and time off, among others (Revolut People API).
Questions to bring to a call
- Which AI features are switched on by default, and which does an admin enable (for example, Performance Insights is enabled in review settings)?
- Is the MCP server available to our workspace yet, and which client would we use?
- Which of our meeting and work tools feed 1:1 summaries and Insights? The stack-fit checker answers most of this in two minutes.
- Data handling and security for AI features: ask on the call and I’ll send the security pack. See also security and access.
See it on your own use case
Bring one review or one report you want to speed up to a 15-minute discovery call.
FAQ
Does the AI write the review for the manager?
It can draft feedback from recorded 1:1s, but the manager chooses whether to keep the draft and can edit it. The docs say AI suggestions may be inaccurate or incomplete and should always be reviewed before submitting. Source.
Can the AI Assistant see data I can’t?
No. The docs say it only sees what you can already see, and only makes changes when you ask it to. Source.
Is the MCP server generally available?
Not yet. It is a Beta feature under a limited release, coming soon to all customers. Source.
Which AI tools can connect through MCP?
The docs have setup guides for Claude Code, Codex App, Cursor and OpenCode. Source.
Where does Performance Insights get its data?
From goals and epics, Jira, GitHub, Bitbucket, 1:1 meetings, manager survey scores and interviews, within the review cycle dates, matched by email. Source.