I Added a "Failure Library" to My MCP Server. Now My AI Agents Warn Each Other About Crashes Before They Happen.

To prevent AI agents from repeating failing operations and wasting tokens, a failure library was integrated into an MCP server. Initially utilizing a JSON-based cache to log errors and share fixes, the system evolved to use SQLite. This upgrade resolved issues regarding concurrent write corruption and visibility across multiple machines. Now, agents check the SQLite database to avoid known traps or apply verified fixes, ensuring more efficient operation across distributed environments.