Example project · A developer web app extended over sixteen months
mcptest.io: a developer tool extended over sixteen months with ProPR.
An existing web app that has kept shipping through ProPR since July 2025. As of 11 November 2026 it holds 288 merged pull requests, 200 of them written by agents, and the busiest month was the most recent full one, October 2026.
Overview
What it is
mcptest.io is a browser-based inspector, playground, evaluator and public catalog for remote Model Context Protocol (MCP) servers. MCP is the standard that lets AI assistants call outside tools. Paste a server's address and mcptest.io connects to it, lists what it offers and shows the raw traffic. A headless release gate runs the same checks in CI, the automatic tests that run before a release. Try it at mcptest.io.
Where it started
The repository was created in April 2025, and its first pull requests were written by hand. On 16 July 2025 one of ProPR's earliest versions opened pull request #19, which added automatic detection of the two ways a remote MCP server can talk. It was the first of 200 agent pull requests.
The project is still shipping daily. The newest example below merged on 11 November 2026, sixteen months after the first.
How the work ran through ProPR
- Each request became an issue.A bug report, a new check or a guide to write is filed as an issue with the
AIlabel, and 217 issues carry it. - An agent answered each issue with a pull request.Existing code means the agent reads before it writes. ProPR gathers the relevant files and history from the repository, and the agent works on its own branch and sends back a pull request that changes only what the issue asked for. How ProPR gathers context.
- Bigger work ran as epics.An epic is a branch where the pull requests for one feature collect, so a redesign or a new report lands as one reviewed set. New models slotted into the same process as they arrived: Claude Opus 5.5 wrote 58 of the pull requests, GPT-5.6 Sol 42, GPT-6.1 Sol 16, GPT-5.5 15 and Claude Fable 5.1 five, after 62 from ProPR's early naming format.
The numbers, as of 11 November 2026.
Agents that wrote the pull requests: Claude Opus 5.5 (58 pull requests), GPT-5.6 Sol (42), GPT-6.1 Sol (16), GPT-5.5 (15), Claude Fable 5.1 (5), 62 more under the early naming format. The second-busiest month was August 2026 with 55 merged. Counts come from the public repository on GitHub.
Four pull requests from the repository
The titles below are quoted from the public repository, from the first agent change in 2025 to one merged on 11 November 2026. Open one to see the issue it came from, the change it made and the review it went through.
- Feat: Add automatic SSE/HTTP transport detection and display#19
- UI/UX Overhaul: Modernize the Application's Look and Feel#123
- Client compatibility matrix as structured, sourced data, rendered in the guides and the report#382
- Detect non-deterministic tool lists within one run (mixed deploys, stale replicas)#577
A whole-app redesign, a data table rendered in two places and a subtle detection bug took the same path, and all three are pull requests you can read today.
What this shows for your own project
Most projects look like this one: code that exists and needs to keep moving. Write each request as an issue, get back a pull request that changes only what the request needs, review it and merge it, and the next request starts from the code as it is now.
Over sixteen months the models changed five times and the process never had to. The repository shows which agent wrote what, and when, and a pull request from 2025 is as easy to open today as one from last week.