What it does
mathlas is an MCP server that gives an AI agent 12 tools for mathematical work. The tools return data such as candidate theorems, verdicts and checklists, and the agent does the reasoning. Theorem search runs over a dense index of 3,683,428 documents combining dense and Okapi-BM25 retrieval. Other tools identify numeric constants with PSLQ and closed-form matching at 50-digit precision, match integer sequences exactly against a local OEIS copy, verify numeric and Lean 4.31.0 kernel claims, look up formal declaration names through the public Loogle and LeanSearch services, and check Ramanujan-style relations. A FunSearch tool runs in a sandbox.
It suits a user of Claude Code or Cursor who wants the assistant to check a numeric claim, identify an unknown constant or integer sequence, find the Lean/mathlib name of a result, or put verified mathematics into an agent pipeline.
It installs with uvx mathlas-mcp or pip and registers as a stdio server. If the mcp SDK is missing, it falls back to a dependency-free JSON-RPC server. The publisher reports 0 false positives across its benchmark tiers and, in one agent comparison on 18 tasks, 18/18 with the tools against 15/18 without them. These are the publisher's own figures.
Limitations and requirements
Requires uv or pip to install. The OEIS sequence tool needs a local OEIS copy and verify_formal needs a Lean toolchain; without them those tools return a data or toolchain unavailable message. Code is Apache-2.0, but the published corpus carries per-source terms (CC-BY or CC0). The 30 GB embedding matrices and local benchmark slices are not included in the published corpus. Benchmark results are self-reported by the publisher.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
io.github.Archerkattri/mathlas
Airtight math for AI agents: 3.68M-doc theorem search + numeric/Lean verification. No LLM, no key.
v1.5.0As shown on registry.modelcontextprotocol.io when we last read it