search-math

Math Resource Search

Companion to search-wolfram — searches external (non-Wolfram-ecosystem) mathematics resources. Use it whenever a math-research project needs to anchor a concept against authoritative external references.

What you need

1. Topic / keywords — what to search for. Concept ("Lie algebra"), theorem ("Stokes theorem"), object ("Bessel function"), sequence ("Fibonacci numbers"), or named entity ("Galois theory").
2. Project directory (optional) — for logging results into Wiki/ if it exists.

Kernel execution (license-aware)

The search scripts below each spawn a fresh wolframscript kernel, which consumes a license seat. Before running them, check headroom on the AgentTools MCP (no extra seat):

With[{free = $MaxLicenseProcesses - $LicenseProcesses}, free]

If free <= 0, do not spawn wolframscript — run the equivalent Import[...] query through mcp__Wolfram__WolframLanguageEvaluator, which reuses the one persistent kernel. The .wls scripts remain the path when the MCP is unavailable.

Sources

Run scripts in parallel. Skip any source that is clearly irrelevant (e.g. skip OEIS for a non-discrete topic).

1. MathWorld

Wolfram MathWorld — large encyclopedia, classical math through research math.

wolframscript -f "${CLAUDE_PLUGIN_ROOT}/scripts/search_mathworld.wls" <keyword1> <keyword2> ...

Returns JSON array of {name, url, snippet}. The script scrapes the search page for entry links matching mathworld.wolfram.com/<EntryName>.html. Names are derived from CamelCase slugs.

2. nLab

Category-theoretic and higher-structures wiki. Best for category theory, homotopy theory, algebraic geometry, mathematical physics.

wolframscript -f "${CLAUDE_PLUGIN_ROOT}/scripts/search_nlab.wls" <keyword1> <keyword2> ...

Returns JSON array of {name, url, snippet}. Filters out archival "YYYY Month changes" pages.

3. OEIS

Online Encyclopedia of Integer Sequences — only relevant when an integer sequence appears.

wolframscript -f "${CLAUDE_PLUGIN_ROOT}/scripts/search_oeis.wls" <keyword1> <keyword2> ...

Returns JSON array of {anum, name, sequence, url}. anum is the A-number (e.g. A000045). sequence is the first ~120 chars of the data row.

You can also query by first terms:

wolframscript -f "${CLAUDE_PLUGIN_ROOT}/scripts/search_oeis.wls" 1 1 2 3 5 8 13 21

4. DLMF

NIST Digital Library of Mathematical Functions — authoritative for special functions, asymptotics, identities.

wolframscript -f "${CLAUDE_PLUGIN_ROOT}/scripts/search_dlmf.wls" <keyword1> <keyword2> ...

Returns JSON array of {name, url, location}. location is the DLMF locator (e.g. 10.2#E5 is chapter 10, section 2, equation 5).

5. Wikipedia math

Filtered Wikipedia search — uses the MediaWiki search API and ranks results by math-keyword overlap and presence of LaTeX markers in snippets.

wolframscript -f "${CLAUDE_PLUGIN_ROOT}/scripts/search_wikipedia_math.wls" <keyword1> <keyword2> ...

Returns JSON array of {title, url, snippet}.

After searching

Present a single combined summary, grouped by source:

Math Resource Search: <topic>

MathWorld:   <n> entries
- <name> — <url>

nLab:        <n> entries
- <name> — <url>

OEIS:        <n> sequences
- A000045 — Fibonacci numbers — <url>

DLMF:        <n> formulas
- 10.2#E5 — <url>

Wikipedia:   <n> articles
- <title> — <url>

Then ask the user which to keep. For each kept result:

1. If Wiki/ exists: invoke add-resource with the URL — add-resource will recognise the URL pattern and create the right kind of wiki article (MathWorld / nLab / OEIS / DLMF / Wikipedia).
2. If no Wiki/: just print the URLs as references the user can keep.

Integration with other skills

• Pair with search-wolfram when the project also touches Wolfram-ecosystem material.
• Pair with cite to immediately produce a BibTeX entry for any DOI/arXiv ID surfaced (e.g. references inside a MathWorld or Wikipedia article).
• Pair with add-resource to commit kept results to Wiki/Resources/.

When NOT to use

• Pure Wolfram-ecosystem questions → use search-wolfram.
• Looking for a specific paper (arXiv / DOI / title) → use the existing arXiv MCP and cite.
• Reading an already-added paper → use mcp__arxiv-latex-mcp or mcp__arxiv__read_paper.