finsliparn

Finsliparn is a Bun-native MCP Server that orchestrates a refinement loop for code tasks. It runs your initial implementation, executes your test suite, and then feeds structured feedback back to the AI to guide subsequent iterations. The result is a multi-attempt, test-validated approach to coding that prioritizes passing tests and minimizing complexity. You can interact with it via Claude Code or the GitHub Copilot CLI, enabling an integrated workflow where the AI suggests changes, your tests validate them, and the cycle repeats until all tests pass. The system uses a dedicated directive.md as the single source of truth for what the AI should do next, and each iteration runs in an isolated worktree to keep the main branch clean.

Prerequisites:

• Bun v1.3+ (https://bun.sh)
• Git

Installation (development):

1. Clone the repository git clone https://github.com/jgabor/finsliparn.git cd finsliparn

2. Install dependencies with Bun bun install

3. Start the MCP server bun run start

4. (Optional) Configure editor integrations:

   • Claude Code: follow the integration guide once available
   • GitHub Copilot CLI: use the provided commands to start the agent

Notes:

• Ensure you have a compatible Node/Bun environment and that your project has a valid package.json if you extend the start script.

Tips and common issues:

• The server relies on a directive.md as the primary instruction source. Ensure it is present and up-to-date in your workspace.
• Each iteration uses a separate worktree; commit results only after tests pass to maintain a clean main branch.
• If tests fail due to environment issues, verify that Bun is correctly installed and that your test suite is runnable from the project root.
• For debugging, set FINSLIPARN_LOG_LEVEL to debug and inspect logs for test execution details.
• If you plan to run this with Claude Code or Copilot CLI, ensure those connectors are configured to point at the running MCP server endpoint.