
Mathlas
Plug this into Claude Desktop or Cursor and your AI gets twelve tools for verifiable math: search across a 3.68 million document index of theorems and arXiv papers, identify numeric constants via PSLQ at 50-digit precision, match integer sequences against a local OEIS copy, verify formal proofs through Lean's typechecker, and decompose theorem applicability into atomic precondition checklists. No LLM calls happen inside the server. It returns raw data for the AI to reason over. The headline is zero false positives: numeric verification re-evaluates at high precision, sequence matching is exact term lookup, formal verification runs the actual Lean kernel. Install with uvx, no API key needed. The retrieval tier gracefully degrades when optional datasets are missing.
- GitHub stars
- 12
- tools
- 12
- transport
- STDIO
- env vars
- 2
- last updated
- Jul 2, 2026
Tools
12 toolsPublic tool metadata for what this MCP can expose to an agent.
search_existing_mathquery → ranked results from the 3.68M-doc dense + BM25 + RRF index
query → ranked results from the 3.68M-doc dense + BM25 + RRF index
No parameter schema in public metadata yet.
identify_constanta real value → known closed form + provenance (50-digit re-eval)
a real value → known closed form + provenance (50-digit re-eval)
No parameter schema in public metadata yet.
verify_numericdigit-agreement verdict — different engine, higher precision
digit-agreement verdict — different engine, higher precision
No parameter schema in public metadata yet.
verify_formalruns the real Lean kernel — typecheck a snippet, or pass proof to kernel-check a full Lean 4 proof
runs the real Lean kernel — typecheck a snippet, or pass proof to kernel-check a full Lean 4 proof
No parameter schema in public metadata yet.
search_formal_mathmathlib declaration names + types via the public Loogle (pattern/type) + LeanSearch (natural language) services, provenance-labeled
mathlib declaration names + types via the public Loogle (pattern/type) + LeanSearch (natural language) services, provenance-labeled
No parameter schema in public metadata yet.
identify_sequenceinteger sequence → matching OEIS entries (exact term-match)
integer sequence → matching OEIS entries (exact term-match)
No parameter schema in public metadata yet.
applicability_checklistresult's hypotheses as an atomic checklist for the AI to mark
result's hypotheses as an atomic checklist for the AI to mark
No parameter schema in public metadata yet.
mapping_scaffoldneeds↔guarantees questions + fill-in template
needs↔guarantees questions + fill-in template
No parameter schema in public metadata yet.
conjecture_relationRamanujan Machine: PSLQ over rich basis + CF/recurrence conjectures
Ramanujan Machine: PSLQ over rich basis + CF/recurrence conjectures
No parameter schema in public metadata yet.
funsearchFunSearch harness in one tool — action=evaluate (sandbox-score an AI-written program), register (MAP-Elites DB), status (best + few-shot)
FunSearch harness in one tool — action=evaluate (sandbox-score an AI-written program), register (MAP-Elites DB), status (best + few-shot)
No parameter schema in public metadata yet.
search_directiveweb-search plan: arXiv queries + sub-fields + which tools to run
web-search plan: arXiv queries + sub-fields + which tools to run
No parameter schema in public metadata yet.
add_findingingest a web-found result into the live corpus
ingest a web-found result into the live corpus
No parameter schema in public metadata yet.
Sponsors
Configuration
MATHLAS_SEEDSet to 1 to force the lightweight built-in seed corpus and never load the multi-GB prebuilt index (fast cold start).
MATHLAS_INDEXPath to a prebuilt index .npz to serve for search_existing_math (optional; the seed corpus is used when absent).
Readme
Details
- Repository
- github.com/archerkattri/mathlas
- Categories
- AI & LLM Tools, Search & Web Crawling
- Package
mathlas-mcp- Transport
- STDIO
- Registry
- MCP registry, active