A Model Context Protocol server that hands your LLM a real Borland Turbo C 3.0 toolchain — compile, run, lint, benchmark, disassemble and explain errors on genuine 16-bit DOS. Real output. Real exit 0.
Most "run my code" tools stop at stdout. This one gives your model a whole workshop — and speaks its language: structured, line-numbered results.
Every compile is parsed into error/warning objects with file + line, so the AI fixes code surgically instead of guessing.
A heuristic linter flags gets(), scanf missing &, = in if, brace mismatches and float-I/O traps — instantly.
A built-in knowledge base turns cryptic Turbo C messages into plain-English cause + concrete fix.
16-bit binaries can't run on 64-bit Windows — the bundled DOSBox bridge makes compile and run work anywhere.
Chain them: analyze_code → compile_check → explain_error → compile_and_run → benchmark.
Node ≥ 18 and Turbo C at C:\TURBOC3. That's it.
# clone, install, build
git clone https://github.com/BBSRguy/turboc-mcp-server.git
cd turboc-mcp-server
npm install
npm run build// claude_desktop_config.json
{
"mcpServers": {
"turboc": {
"command": "node",
"args": ["C:/path/to/turboc-mcp-server/build/server.js"],
"env": { "TURBOC_ROOT": "C:\\TURBOC3" }
}
}
}# one command
claude mcp add turboc -- node C:/path/to/turboc-mcp-server/build/server.jsOn 64-bit Windows? Add the DOSBox bridge → docs/DOSBOX.md. Then ask your AI to run environment_doctor.
Ask: "run the prime-sieve example" — the AI compiles & runs it on real Turbo C.