Anyone who uses an AI coding assistant on a real multi-file project hits the same three walls: the AI changes a function in File A and breaks a caller in File B, follows an outdated design note after the code has changed, or claims “all tests pass” without running them.
Writing longer system prompts doesn’t fix this, because the model forgets them as the chat grows. Resonant is an MCP server that plugs directly into Antigravity, Cursor, and Claude Code to give your AI three deterministic tools that run in milliseconds.
Before your AI saves a file, Resonant inspects the modified function, tests boundary inputs in memory (like empty lists [] or 0), and checks for leaked URL keys or in-place list mutations—catching crashes even when no unit test exists yet.
Every project note or learned bug rule stored in Resonant’s local SQLite database is linked to the SHA-256 fingerprint of the actual code function. The moment you edit that function later, outdated notes about it expire automatically.
Instead of reading whole files line-by-line on every prompt, your AI asks Resonant for a symbol and gets back the exact function signature, the files that call it, its Git co-change partners, and the tests that cover it.
You don’t need to install anything or share your codebase to try it. Add our live remote demo endpoint (https://resonantexo.com/api/mcp) to your mcp_config.json in Antigravity, Cursor, Claude Code, or Windsurf, then ask your AI to run resonant_verify_cegar on any function you are writing:
{
"mcpServers": {
"resonant-demo": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://resonantexo.com/api/mcp"]
}
}
}
Want the full local resident version (tb_mcp_server.py) with local SQLite memory and Git history indexing on your machine? Email dev@thunderbearstudios.org.
Want to see the exact output the MCP server returns before adding it to your editor? Pick one of the three snippets below—or paste your own Python/JS function—and click Run MCP Check. Every number below is measured directly on the text you submit.