An open-source MCP server built by autonomous AI developers: the numbers

Description and use

Over 42 days, a two‑person team built the open‑source MCP server mcp4mail using autonomous AI developers for all coding, testing, CI, and merges. They completed 182 tasks, 146 PRs, logged 140 hours, and achieved 90 % first‑run CI pass rate.

Problem

How to build and maintain an open‑source product with minimal human intervention while ensuring quality and reliability.

Outcome

In 42 days, 182 tasks were finished, 146 PRs merged, 140 hours logged, and 90 % of PRs passed CI on first run. The system handled CI failures automatically and required human intervention only for planning and deployment.

Workflow

• 1. Create a task with brief, context, acceptance criteria and add to queue
• 2. Runner picks task, selects model (Claude Code) and starts coding CLI
• 3. Coding CLI writes code, tests, runs project tests, opens PR linked to task
• 4. CI runs lint, security scan, unit and system tests; if any fail, PR stays open
• 5. If CI passes, PR is merged after linear history and status checks
• 6. Deployed changes are picked up by hosted instance daily around 18:00 Prague time
• 7. Runner monitors progress, respects daily quota, logs, and handles failures by creating urgent tasks or retrying

Evidence

I run a two-person software company in the Czech Republic. For the last six weeks I have been running an experiment in public: an open-source product where the code, the tests, the pull requests and the merges are done by AI developers, and a person steps in only where a person is actually needed.

Source

The product is called mcp4mail. This article is the story of how it was built, with the numbers, and with links that let you check every one of them yourself.

Source

As of 30 October 2026, from the live page: Days of development 42, Tasks finished 182, Pull requests merged 146, Hours of work logged 140, Pull requests that passed CI on the first run 90 %

Source

The runner is a single binary that runs on one of our own machines. It picks the next task that is ready, triages it, chooses a model, and starts a coding CLI.

Source