Tools for AI: mcp

• It runs on your machine and talks over standard input and output: your code never leaves the machine.
• Add it to an assistant as an MCP server whose command is kavach mcp.
• Inside a project, answers come from an index kept in .kavach/index, so they are quick.
• A run or test that is still going after 30 seconds, or the call's own time_limit, is stopped at its next loop round or call, with exit code 124, and the answer says so. A model's code that never ends can't hold the server. tested

The site also serves llms.txt: a short guide to the whole language that an assistant reads once.