Skip to main content

MCP OAuth, bulk repository ingestion, and admin fixes

· 4 min read
Venue.sh Team
Venue.sh Team
Venue.sh

OAuth-gated MCP servers work from the Assistant now, and there is a My connections page to manage them. Repository ingestion moved out of the dashboard into backend workers that can process many repos in one run. Last Signed In, Active devices, and Docs Hub had bugs worth fixing in this release too.

OAuth for MCP servers​

Some MCP servers will not run a tool until you sign in with the provider. Venue.sh reads the server's protected-resource metadata on a 401, registers an OAuth client for you, and lets each user connect their own account. No client ID or secret to paste in.

  • Automatic discovery - Register a URL-based MCP server and Venue.sh handles dynamic client registration when OAuth is required.
  • Inline connect in chat - The Assistant calls a tool you are not authorized for yet? An inline card opens a popup. Authorize there and the tool call picks up where it left off. You do not resend the message.
  • Reconnect when expired - Expired tokens get Reconnect your account, not a silent failure.
  • My connections - Settings > My connections lists every OAuth MCP server you linked, its status, and when access expires. Connect, reconnect, or disconnect from one screen.

Once a server is registered and you are connected, try:

  • "What Jira issues are assigned to me this sprint?"
  • "Search Confluence for our deployment runbook."

See MCP Servers to register servers and Assistant for chat workflows.

Bulk repository ingestion​

Ingestion, embeddings, and summaries no longer run inside dashboard HTTP requests. Backend workers own the job queue now. They retry failed repos, cap concurrency, and keep going when one repository fails.

  • Bulk runs - Kick off ingestion for many repositories in one job. Watch overall progress while each repo succeeds or fails on its own.
  • Skip unchanged repos - If the content digest matches the last run, Venue.sh marks the job skipped. Force a re-ingest when you want a fresh index anyway.
  • Background workers - GitIngest jobs batch by Git connection, credentials stay isolated per batch, and results land in storage so search, summaries, and Skills discovery keep working while the queue drains.

Import and entity updates still trigger ingestion the same way as before. See AI Settings. The difference is you can queue hundreds of repos and close the tab.

Users table fixes​

Last Signed In was blank for everyone. The API still returned an empty string left over from a previous migration. It now reads each user's most recent session in the active organization.

Sorting by Joined or Last Signed In also did nothing before. Both columns reorder the table now.

See Users and Teams for inviting and managing people.

Other fixes​

  • Active devices - Settings > Account > Active devices listed nothing after your session passed Better Auth's 24-hour freshness check, even though you were still signed in for a week. The page reads unexpired sessions from the database now. If the fetch fails, you get an error instead of an empty section.
  • Docs Hub titles and errors - TechDocs pages no longer flash an opaque database ID as the title while loading. When docs are missing, the message says whether the entity was not found or the entity exists but has no TechDocs content configured yet.

See Also​

  • MCP Servers - Register and inspect MCP servers in your organization.
  • Assistant - Use the Assistant with catalog data, plugins, and connected MCP servers.
  • AI Settings - Enable repository ingestion and general AI usage.
  • Importing Repositories - Import repositories into the catalog.
  • Skills - Browse agent skills discovered during ingestion.
  • Settings - Account, API keys, and active devices.
  • Users and Teams - Manage users, roles, and invitations.
  • TechDocs - Publish and browse documentation from catalog Components.