Delegate from your chat assistant.

Connect Claude or ChatGPT to your instance's MCP server and hand off work in the conversation you are already having. ProPR runs the agent on your server, opens the PR in GitHub, and answers the chat when you ask how it went.

Describe
Run isolated
Review in GitHub

Feature overview

You start with
A ProPR instance with MCP enabled, and Claude or ChatGPT
You use
create_task · create_plan · get_operation · get_pull_request_discussion
Takes
One sign-in, then one request per task

Connect the assistant

Add https://<instance>/api/mcp as a custom connector in Claude, or as an app in ChatGPT's developer mode. Both reach your instance from their own cloud, so it needs a public HTTPS address.

Sign in through your instance in the browser and pick the repositories and permissions the assistant gets. Optional scopes start unchecked.

The MCP consent screen’s repository picker: two of four Northwind repositories selected, the starred one first, with Select all, Clear and a filter box
The consent screen grants access repository by repository.

Hand off a task

Describe the change. The assistant calls create_task, which opens a GitHub issue and starts implementation straight away, with the model you name. Ultrafix and auto-merge run only when you ask for them.

Add retry with backoff to the webhook sender in acme/payments. Use Claude Opus 5.5 and run Ultrafix to 8/10.

Starting work needs the execute scope. The default ceiling allows read, plan and review, so an administrator raises it first.

The MCP scope ceiling: read, plan, publish, execute and review allowed; merge, deploy and manage left off
The administrator's ceiling bounds every grant.

Or plan it first

For a larger change, create_plan drafts a plan and generate_plan fills in the tasks. Refine it in the chat, then publish_plan turns it into GitHub issues and implement_plan starts the issues you pick with the models you choose.

Plan rate limiting for orders-api. Show me the tasks before you publish anything, then start the first two with GPT-6 Astra.

Follow the progress

Every request returns a receipt. Ask how it is going and the assistant reads get_operation or get_task_submission, which report each stage from issue to finished PR. get_work_overview joins running tasks with their PRs, reviews and checks.

Close the chat and the work keeps running. A new conversation picks it up again from the operation ID or the PR number.

Read the PR discussion

get_pull_request_discussion brings the PR's comments into the chat, ProPR reviews and their F# findings included. Pick the findings worth fixing and fix_review_findings applies exactly those as a follow-up commit.

Summarize the review on acme/payments#57 and apply F2. Leave the suggestions.

The chat asks. Your instance decides.

ProPR checks every call against the grant, the administrator's ceiling and your own GitHub permissions. Merging takes its own merge scope and the exact head commit. Keys and tokens belong in browser sign-in; keep them out of the conversation.

Administrators see each call in the MCP Log. Follow the chat assistant guide to connect a client and the MCP reference for the full tool list.

The MCP Log: each tool call with its connected app, outcome and duration; one call is denied with FORBIDDEN_REPOSITORY
A call to a repository outside the grant is logged as denied.

What you get

Tasks and plans started from a conversation, running on your server, landing as reviewed PRs in GitHub, with their progress one question away.

MCP from chat → All playbooks