The-40-Thieves/codecalc

Overview

Provides 52 MCP tools across 31 languages for AI models to execute and evaluate code. Uses a Rust sandbox for safe execution, with verdicts, sessions, artifacts, exact arithmetic, and verified translation or optimization. No LLM, gateway, or telemetry. The calling model interacts directly. Apache-2.0 licensed.

Best for

Best for
Developers building AI agents that need safe, multi-language code execution

Use cases

• Let an AI model run and test generated code in a sandbox
• Verify code translation between languages with exact results
• Perform exact arithmetic or optimization checks during model reasoning

How to use

Install

uvx 'codecalc[full]' setup            # prints what it would do — nothing on disk changes

Tools exposed

• install_package
• calc_exact
• compare_threshold
• calc_stats
• collision_probability
• data_sizes
• human_duration
• epoch_time
• radix_convert
• float_repr
• algebraic_equiv
• list_languages
• list_execution_providers
• execute_code
• execute_code_stream
• trace_execution
• branch_reachability
• run_submit
• run_inspect
• run_cancel

Tested with

Claude Desktop, Claude Code, Cursor, VS Code, ChatGPT

Notes

Provides 52 MCP tools across 31 languages for AI models to execute and evaluate code. Uses a Rust sandbox for safe execution, with verdicts, sessions, artifacts, exact arithmetic, and verified translation or optimization. No LLM, gateway, or telemetry. The calling model interacts directly. Apache-2.0 licensed.

3 stars on GitHub. Last updated 2026-09-08. Licensed Apache-2.0.

Use cases

• Let an AI model run and test generated code in a sandbox
• Verify code translation between languages with exact results
• Perform exact arithmetic or optimization checks during model reasoning

Pros

• Broad language coverage with 31 languages and 52 tools
• Sandboxed Rust execution for safety
• No telemetry or gateway, keeping the model as the direct caller

Cons

• Very low community traction with only 3 stars
• Requires an MCP environment and direct model integration
• No bundled LLM or gateway, so the caller must handle orchestration

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.