Your MCP servers and skills, inside every task run you choose.

An extension profile bundles the MCP servers and skills an agent should have: your issue tracker, your internal docs, your team's review checklist. Pick a profile on New Task or an automation, and the run starts with exactly that set.

Gather context
Execute
Refine the PR

Feature overview

MCP servers, tool by tool

Add a remote HTTPS MCP server, discover its tools, and tick the ones the agent may call. Every other tool stays out of reach.

Skills, pinned to a commit

Import a skill from a GitHub repository or an archive. ProPR pins the exact commit and content digest, then mounts the skill read-only in the task container.

Versioned profiles

Every save creates a new revision. A queued run keeps the revision it was given, so an edit never changes work already in flight.

The agent gets the tools. The credentials stay with ProPR.

A server's bearer token is stored encrypted on your instance and never sent to the task container. The agent reaches each server through a per-run gateway that ProPR revokes when the run ends.

  • The gateway forwards only the listed tools, with bounded requests and responses.
  • It refuses redirects and private, loopback and cloud-metadata addresses.
  • Restricted network mode still applies: a profile never opens a host the run's allowlist doesn't.
  • Each run's gateway traffic is recorded on the task timeline.
An agent extension profile: the runbooks MCP server over HTTPS with two read-only tools and a write-only bearer token, plus the incident-runbooks skill pinned to a commit with its SHA-256 digest
A profile with a runbooks MCP server, two allowed tools, and a skill pinned to a commit.

Write access is a separate, explicit grant.

Mark a server read or write. Write servers stay out of every run until the run is authorized to write: New Task shows an unchecked opt-in, and automations follow their acting step's policy. Report runs never see them.

  • Claude Code and Codex task runs take profiles. Goals run without them.
  • Profiles can be limited to specific agents and repositories.
  • Operators can also install local stdio servers in the agent image and reference their secrets from the server environment.
  • Skill scripts are task input. ProPR never runs them on import.

Where it lives.

Instance administrators manage profiles in Settings → Agent Extensions & Packages. The agent extensions guide covers endpoint rules, the operator catalog, skill imports and backups. Agents run in the same isolated workspace as every other task.