17 read-only tools that turn the official Codeforces API into compact, structured JSON — problem search, practice ladders, rating history, performance analytics, standings and live contests — for Claude, Cursor, Cline and any MCP client. Same tools, same outputs, in Node.js and Python.
Everything a competitive programmer asks about
Find problems by id (158A), title, tag or rating. Multi-word queries use AND semantics, so dp 1500 means exactly that.
Per-contest deltas, best rating, biggest swing, acceptance rate, verdict mix, favourite languages, strongest tags and a solved-by-rating curve.
Upcoming rounds with hours-to-start, standings windows filtered by handle, post-contest rating changes and per-contest submissions.
No login, no cookies, no code execution, no submissions. get_capabilities states the contract so agents never try.
A serialized throttle, TTL cache and exponential backoff on Call limit exceeded, 429, 5xx and timeouts keep you inside Codeforces' limits.
The 20 MB problemset, 26 MB rated list and full standings are normalized and capped. Submissions are compacted to the fields that matter.
Copy a config, restart your client
Loading config...
17 tools, one contract
Arguments shown in Node.js camelCase; the Python runtime uses snake_case equivalents.
No tools match that search.
Every rating in every response carries its Codeforces rank name.
One pipeline, two runtimes
src/app/Hardened against real API quirks
lang=en is sent.1500 returns 1500-rated problems, not contest 1500.| Node · offline unit | 89 / 89 |
| Node · live API smoke | 21 / 21 |
| Node · MCP over stdio | 86 / 86 |
| Python · unit, mocked API, MCP stdio | 78 / 78 |
| Python · live API | 9 / 9 |
CI runs the offline suites on Node 20 / 22 / 24 and Python 3.10 / 3.12 / 3.13.