Skip to main content
Kontext exposes a hosted Model Context Protocol server at https://mcp.getkontext.io. Any MCP-compatible client — Claude, Codex, Cursor, Continue, Zed, custom agents — can connect to it and gain tools for reading and writing your product data: products, features, releases, feedback, detected problems, actors, contexts, and insights. All calls are scoped to the organization the user picks during authorization and enforced against the scopes they approved.

​
What you get

  • Read tools — list and fetch products, features, releases, feedback, problems, actors, contexts
  • Write tools — submit feedback, create features, releases, actors, contexts
  • Interactive UI resources — tools like get_insights_chart return both raw data and a rendered chart the client displays inline (via the MCP Apps extension)
  • OAuth with scoped permissions — consent screen shows exactly which resources the client asked for; you can deny any scope
  • Full auditability — authorized clients appear in Kontext under Organization Settings → Connected Apps with the scopes they were granted and when they were last used

​
How it compares to the REST API

Both the MCP server and the REST API speak the same underlying endpoints with the same scope model. Pick based on caller shape: MCP gives the human an OAuth dialog to pick the org and scopes. API keys give non-interactive callers a long-lived credential.

​
Managing connected clients

Every MCP client that completes the OAuth flow gets a row in Organization Settings → Connected Apps in the Kontext web app — with the client name, the granted scopes, and the last-used timestamp. Revoking a row immediately invalidates its token; the client has to re-authorize to reconnect.

​
Next steps

  • Setup — wire up Claude, Codex, or any OAuth-capable MCP client
  • Tools — the full tool reference, with arguments and required scopes
  • UI resources — how interactive chart tools render inline in the client