MCP & context
Context is what AI reads when it drafts OKRs, splits them across reports, drafts initiatives and suggests progress. Connect tools over the Model Context Protocol (MCP), crawl docs sites and public pages, or upload files.
MCP connectors
Open Context in the left rail. Presets are grouped by category, and hosted endpoints are pre-filled where the vendor publishes one (marked hosted). Every field stays editable.
| Category | Connectors |
|---|---|
| Engineering | GitHub, GitLab, Sentry, Figma, Azure DevOps |
| Project management | Atlassian (Jira, Confluence), Linear, ClickUp, Asana, monday.com, Notion, Shortcut, Trello |
| Change & incidents | ServiceNow, PagerDuty, Jira Service Management, Opsgenie |
| HR (HRMS) | BambooHR, Workday, HiBob, Rippling, Personio, Deel |
| Docs & knowledge | Google Drive, Slack, any custom MCP server |
For tools without a hosted MCP endpoint, paste the URL of the MCP server your team runs for it.
Sign in to a connector
Each connector connects with OAuth sign-in or an access token.
OAuth sign-in
- OAuth follows the MCP authorization spec. The client discovers the server’s sign-in settings from its
401response, registers itself where the server allows it, and signs in with PKCE in a popup. - Tokens are stored encrypted and refreshed automatically. When a sign-in expires or is revoked, the source shows Reconnect.
- Servers that don’t let apps register themselves need an OAuth app you create. Use this redirect URI:
<OKRAI_URL>/api/mcp/oauth/callback
Access token
Paste a personal access token, API key or service token. It is sent as Authorization: Bearer … by default; some presets use a different header. Tokens are encrypted at rest and sent only to that server.
What OKR.ai reads
- The built-in MCP client reads resources and calls only read-only, argument-free tools, most OKR-relevant first: goals, projects and epics, teams and people, then issues and tickets.
- JSON output is converted to compact markdown and cached on the source in the workspace’s own database, then refreshed daily. Use Refresh to fetch now.
- AI treats connected content as data, never as instructions.
Docs, pages and files
- Docs site: crawl a docs or handbook site.
- Public page: a company site, strategy page or careers page.
- Upload files: PDF (sent to the model as a document), DOCX, Markdown, text, CSV and JSON.
Crawls run behind an SSRF guard that blocks private-network hosts. Set OKRAI_ALLOW_PRIVATE_URLS=1 to allow an internal wiki.
Using context
- From context in My OKRs or Reports drafts OKRs for you and your team. See OKR workflow.
- Suggest from context on a key result estimates its current value.
- Pull, push and initiative drafts use connected context automatically.