Pre-forged tools: landscape and catalog

Status: proposed (16 Oct 2026). Part of the registry design. Two reference entries prove the manifest shape and pass their tests through tools/catalog/run_tests.py with the axlr_run_tool contract (stdin JSON, PATH and HOME only): seat_allocation and business_days. Counts, stars and dates below were read on 16 Oct 2026 with gh, curl and page fetches; what could not be verified says so.

Landscape: what exists and what it teaches🔗

Catalog	Size	What an entry carries	Organisation	Trust
Official MCP registry (schema)	> 6,000 servers (page cap hit; exact total not read)	`server.json`: reverse-DNS `name`, `description`, `version`, `repository`, `packages[]` with `registryType` (npm, pypi, oci, mcpb…), `environmentVariables{required,secret}` and `fileSha256` for bundles	flat; the namespace is the only grouping	namespace ownership (GitHub OAuth, DNS TXT, `.well-known`); metadata only, no code scan
Glama	100,469	tags, licence/quality/maintenance grades A–D, tool list	facets: language, hosting, capability, subject (Developer Tools 23.7k, Search 14.3k, Finance 6.7k, Databases 5.6k, Security 5.0k…), Official/Claimed	letter grades, method unexplained
Smithery	27,172+ MCPs, plus Skills	owner/slug, description, usage count	none visible	a "verified" badge with no stated check
PulseMCP	21,698	classification, weekly visitors	Reference / Official / Community	badges; submissions paused
mcp.so	not shown	config, tools	five top categories plus tags	manual review of config and tools
Composio	"1000+ toolkits"	`TOOLKIT_ACTION` slug, input and output schema; auth lives outside the tool	by service; a `COMPOSIO_SEARCH_TOOLS` meta-tool for runtime discovery	managed OAuth; every tool wraps a SaaS API
strands-agents/tools	50 modules, 1.3k stars	a Python function; docstring and type hints make the schema	flat	repository review
anthropics/skills, Agent Skills spec	17 skills, 182k stars; spec repo 26k	`SKILL.md` frontmatter: `name` (≤ 64, lowercase, equals the directory), `description` (≤ 1024), `license`, `compatibility`, `metadata`, `allowed-tools`; `scripts/`, `references/`, `assets/`	plugin bundles in `marketplace.json`	a git marketplace, no checksums; `skills-ref validate` lints
LangChain community, LlamaHub, CrewAI tools, smolagents	81 / 67 / ~80 / ~10	a class with `name`, `description`, `args_schema`	by vendor	none; PyPI
ToolBench, Gorilla	16,464 APIs / 1,600+	per-tool JSON with `api_list`	49 RapidAPI categories; G1/G2/G3 instruction groups	pass-rate filtering; BFCL leaderboard
npm, PyPI (PEP 691/740), Homebrew `formula.json`	8,662 formulae in one static file	`dist.integrity`, `hashes.sha256`, `bottle.files.<platform>.{url, sha256}`	flat JSON, per-item pages	per-artifact hashes; registry signatures (npm ECDSA, PyPI attestations)

Four lessons the design follows:

Only the MCP registry and the package indexes carry per-artifact hashes; AXLR's index does, with a signature over the whole index (D1, D2).
Nobody execute-tests catalog entries. A CI-run tests.json per tool is AXLR's verification signal and the thing that makes a catalog tool safer than a freshly forged one.
Categories are weak everywhere; Glama's subject facets and RapidAPI's 49 categories are the usable references. AXLR's categories below are by the person's domain, not by vendor.
Almost every catalog entry wraps a SaaS API. Offline, deterministic, non-programming tools are the gap in the market, and they are exactly what a forged-tool runtime with no secrets can run.

Entry schema🔗

A catalog tool is a ForgedTool plus a catalog block (D3):

Field	Meaning
`name`, `description`, `input_schema`, `program`, `args`, `files`	exactly the forged-tool fields; `args` names one of `files`
`catalog.version`	semver
`catalog.category`, `catalog.tags`	one category from the list below; free tags for search
`catalog.license`	SPDX id, MIT or Apache-2.0 compatible
`catalog.requires`	`program_min_version` and `packages[]` (empty in tier 1)
`catalog.network`	`true` when the tool opens a socket
`catalog.example`	one input and the output it produces, shown on the site and in `axlr_catalog describe`
`catalog.source`	where the method or idea comes from
`tests.json`	at least three cases: `{name, input, expect: {exit, stdout_json
`README.md`	what it does, the method, its limits; for health, law and tax tools, the disclaimer the output also prints

Tiers: 1 stdlib only (Python 3.10+, bash with coreutils, Node 22 without packages), no network: installed with a single card. 2 needs a package (requires.packages): the card warns when the import fails. 3 needs the network, no key: works without the sandbox or with its network allowed. 4 needs an API key: not in the catalog until the secrets mechanism exists.

Conventions: every tool prints one JSON object; a usage error exits 2 with {"error": …}; numbers that matter are exact (fractions, decimal); nothing reads files outside the workspace unless the input names them; a tool whose answer depends on a dated table (holidays, CPI, tax brackets) carries the table's date in its output and its version.

Catalog🔗

Tier 1 unless marked. "Source" names the implementation to learn from or "original".

Politics and civics🔗

Tool	Description	Source
`seat_allocation`	D'Hondt, Sainte-Laguë, modified Sainte-Laguë, Hare; threshold; award order and ties (reference entry)	original
`seat_allocation_droop`	Droop and Imperiali quotas, largest remainder	original
`vote_condorcet`	pairwise matrix, Condorcet winner, Schulze and Copeland	CIVS ideas
`vote_irv`	instant-runoff rounds with tie handling and exhausted ballots	original
`vote_borda`	Borda and Dowdall scores	original
`vote_stv`	single transferable vote (Gregory method)	original
`apportionment`	Huntington-Hill, Webster, Jefferson apportionment of seats among districts by population	original
`gerrymander_metrics`	efficiency gap, Polsby-Popper and Reock compactness from district vote and geometry data	original
`poll_margin_error`	sample size, margin of error, lead significance	original
`legislative_calendar`	term deadlines from a bill's dates with a session calendar given in the input	original
`transparency_request_draft`	access-to-information request template per jurisdiction (Spain Ley 19/2013, EU 1049/2001, US FOIA, UK FOIA), with "not legal advice"	original
`budget_breakdown`	per-capita, per-programme and year-over-year breakdown of a public budget CSV	original

Administration and office🔗

Tool	Description	Source
`business_days`	count, add, list working days with weekends and holidays given in the input (reference entry)	original
`holiday_table`	holidays for a country and region and year, from a bundled dated table (tier 2 with `holidays`)	vacanza/holidays
`invoice_calc`	lines, discounts, VAT per rate, withholding, totals in `Decimal`	original
`iban_validate`	IBAN checksum, country length, BBAN split	python-stdnum
`es_tax_id_validate`	Spanish NIF, NIE and CIF check letter	python-stdnum
`eu_vat_validate`	EU VAT number syntax and checksum per country	python-stdnum
`ics_generate`	RFC 5545 event or recurring event, with alarms and attendees	icalendar
`meeting_planner`	overlap of working hours across time zones (`zoneinfo`)	original
`mail_merge`	CSV plus a text template to many documents	original
`doc_template_fill`	fill `{{placeholders}}` in text or `.docx` (zip and XML, no package)	docxtpl ideas
`date_parse`	natural-language dates and durations (tier 2 with `dateparser`; a stdlib subset for ISO and common forms)	dateparser
`expense_split`	split shared expenses and settle with the fewest transfers	original
`letter_formal`	formal letter skeleton per language and register with the right salutations and closings	original

Mathematics and statistics🔗

Tool	Description	Source
`exact_calc`	expression evaluator over `Fraction` and `Decimal` with an AST whitelist (no `eval`)	original
`stats_describe`	mean, median, SD, quartiles, confidence interval, correlation, linear regression	stdlib `statistics`
`prob_dist`	binomial, Poisson, normal, hypergeometric pmf and cdf; combinatorics	stdlib
`hypothesis_test`	Welch t, paired t, chi-square, Fisher exact, with p-values (exact where stdlib allows; tier 2 with `scipy` otherwise)	scipy
`number_theory`	factorisation, gcd/lcm, primality (Miller-Rabin), modular inverse, totient, CRT	original
`matrix_ops`	determinant, inverse, rank, solve over `Fraction`	original
`unit_convert`	conversion with dimensional check from a bundled table (tier 2 with `pint`)	pint
`base_convert`	bases 2–36, Roman numerals, two's complement	original
`sequence_tools`	OEIS-style detection of arithmetic, geometric and polynomial sequences; next terms	original
`sympy_solve`	solve, simplify, differentiate, integrate, limit, series (tier 2 with `sympy`)	sympy
`sampling_plan`	sample size for proportions and means, stratified allocation	original
`rounding_report`	significant figures, uncertainty propagation	original

Research and science🔗

Tool	Description	Source
`citation_format`	BibTeX or CSL-JSON to APA 7, MLA 9, Chicago, IEEE, Vancouver	CSL styles
`bibtex_tools`	parse, deduplicate, normalise keys, find missing fields	original
`doi_isbn_validate`	DOI syntax, ISBN-10/13 checksum and conversion, ISSN	original
`doi_lookup`	metadata from Crossref (tier 3)	Crossref REST
`arxiv_lookup`	search and fetch arXiv metadata (tier 3)	arXiv API
`chem_formula`	parse `Ca(OH)2`, molar mass, mass percent, from a bundled element table	original
`physics_constants`	CODATA 2022 constants with units and uncertainty	NIST
`dna_tools`	reverse complement, GC content, translation, ORF search	Biopython ideas
`experiment_power`	statistical power and sample size for two-group designs	original
`data_dictionary`	describe a dataset's columns into a data dictionary skeleton	original
`prisma_counts`	PRISMA flow numbers from screening decisions	original

Writing and creativity🔗

Tool	Description	Source
`readability`	Flesch, Flesch-Kincaid, Gunning Fog, SMOG; Spanish via Fernández-Huerta and Szigriszt-Pazos	textstat
`text_report`	word and sentence counts, frequency, n-grams, reading time, longest sentences	original
`style_lint`	passive voice, filler words, adverbs, repeated words, sentence-length spread (heuristic, English and Spanish)	original
`syllables_rhyme`	syllable count, stress, rhyme scheme check (Spanish rules exact; English heuristic)	original
`meter_check`	scan verse against a meter (endecasílabo, alejandrino, iambic pentameter)	original
`name_generator`	seeded names for characters, places and products, by culture and style	original
`story_structure_check`	outline beats against three-act, Save the Cat, hero's journey	original
`md_convert`	Markdown to HTML, plain text or a deck outline	original
`latex_table`	CSV to a LaTeX table with escaping and alignment	original
`slug_title_case`	slugs, title case per style guide, smart quotes	original
`crossword_fill`	fill a crossword grid from a word list	original
`anagram_wordplay`	anagrams, palindromes, lipograms from a bundled word list	original

Programming and devops🔗

Tool	Description	Source
`semver_tools`	parse, compare, bump, match ranges	semver.org
`cron_explain`	explain a cron expression and list the next N runs	original
`regex_test`	run a pattern over text with a timeout, show groups	original
`json_schema_check`	validate against a JSON Schema subset (tier 2 with `jsonschema` for the full spec)	jsonschema
`diff_text`	unified and word diff	stdlib `difflib`
`dockerfile_lint`	latest tag, root user, apt cache, secrets in `ENV`	hadolint ideas
`license_detect`	SPDX licence from a LICENSE text by similarity	SPDX list
`gitignore_gen`	`.gitignore` from named stacks	github/gitignore
`env_example`	`.env.example` from a `.env` with values blanked	original
`openapi_outline`	endpoints, methods and schemas from an OpenAPI file	original
`log_summarise`	group log lines by template, count, first and last time	original

Data and files🔗

Tool	Description	Source
`csv_stats`	per-column type, nulls, min, max, mean, distinct	csvkit
`csv_query`	SQL over CSV files through in-memory `sqlite3`	datasette, miller
`csv_json_convert`	both ways, nesting by dotted keys	csvkit
`xlsx_read`	sheets to CSV through `zipfile` and XML	original
`json_diff`	structural diff with paths	original
`file_hash_tree`	SHA-256 manifest of a directory and verify	original
`encoding_detect`	detect and convert text encodings (tier 2 with `charset_normalizer`)	charset_normalizer
`format_convert`	JSON, TOML (`tomllib`), INI, YAML subset	stdlib
`dedupe_sort`	deduplicate and natural-sort lines	original
`table_pivot`	pivot and group-by over a CSV	original

Finance and economics🔗

Tool	Description	Source
`npv_irr`	NPV, IRR (Newton with bisection fallback), MIRR, payback	numpy-financial
`loan_amortization`	payment and table for French, German and American systems; early repayment	original
`compound_interest`	future and present value, effective rates, rule of 72	original
`inflation_adjust`	CPI adjustment from a bundled dated table (Spain INE, US CPI-U, euro area HICP)	original
`break_even`	break-even, contribution margin, sensitivity	original
`portfolio_stats`	return, volatility, Sharpe, drawdown from a price CSV	original
`tax_bracket_estimate`	bracket-based estimate from a yearly table given in the input or bundled with its year (IRPF, federal income tax), with a disclaimer	original
`fx_convert`	ECB reference rates (tier 3)	ECB
`depreciation`	straight-line, declining balance, sum-of-years tables	original
`cashflow_forecast`	monthly cash-flow projection from recurring items	original

Law and compliance🔗

Tool	Description	Source
`deadline_calc`	calendar versus working-day terms, from a jurisdiction rule given in the input (e.g. "10 hábiles, agosto inhábil"); never guesses the rule	original
`clause_checklist`	checklist per contract type (NDA, services, lease, employment)	original
`gdpr_ropa_template`	Article 30 record of processing skeleton	original
`dpia_screening`	DPIA need screening from the EDPB criteria	original
`license_compat`	compatibility matrix of open-source licences	original
`retention_schedule`	retention periods by record type from a table given in the input	original
`legal_citation_format`	BOE, EUR-Lex, ECLI and US case citation formats	original
`consent_text_check`	consent and privacy notice completeness against GDPR Articles 13 and 14	original

Education🔗

Tool	Description	Source
`quiz_cloze`	cloze questions from a text, with answer key	original
`quiz_mcq_scaffold`	multiple-choice scaffolds with distractors from the same text	original
`srs_schedule`	SM-2 next interval (tier 2 with `fsrs` for FSRS)	py-fsrs
`rubric_grade`	weighted rubric scoring and feedback report	original
`grade_stats`	class statistics, curves, item difficulty and discrimination	original
`flashcards_export`	CSV to Anki-importable TSV	original
`timeline_builder`	events to a Markdown or ASCII timeline	original
`lesson_plan_template`	lesson plan skeleton by level and duration (Bloom verbs)	original
`bloom_classify`	classify learning objectives by Bloom level (verb table)	original

Language🔗

Tool	Description	Source
`tokenize_text`	Unicode-aware words, sentences, counts	original
`unicode_tools`	normalise, strip accents, code points, confusables	stdlib `unicodedata`
`transliterate`	Cyrillic, Greek, Arabic (table-based)	original
`es_verb_conjugate`	regular Spanish conjugation plus bundled irregulars	mlconjug3 ideas
`hangul_romanize`	Revised Romanization from Hangul decomposition	hangul-romanization
`kana_romanize`	Hepburn for kana (kanji needs `pykakasi`, tier 2)	pykakasi
`pinyin_convert`	tier 2 with `pypinyin`	python-pinyin
`language_detect`	character n-gram heuristic (tier 2 with `lingua`)	lingua-py
`number_to_words`	numbers, ordinals and currency to words (es, en, fr, de, pt)	original
`gender_agreement_es`	agreement check for Spanish noun phrases (heuristic)	original

Geography and time🔗

Tool	Description	Source
`haversine`	great-circle distance and bearing, batch	original
`tz_convert`	conversions, DST transitions, offsets at a date (`zoneinfo`)	stdlib
`sun_times`	sunrise, sunset, twilight by the NOAA algorithm	astral
`coord_convert`	decimal degrees, DMS, UTM, geohash, what3words-style not included	original
`calendar_convert`	Gregorian, ISO week, Julian day, Hebrew, Islamic	original
`iso_country_lookup`	ISO 3166, currency, calling code, time zones	original
`moon_phase`	phase and illumination	original
`geojson_tools`	bounding box, centroid, area of GeoJSON polygons	original

Health (deterministic, every output carries a disclaimer)🔗

Tool	Description	Source
`bmi_bsa`	BMI with WHO categories; BSA (Mosteller, DuBois)	original
`egfr_calc`	CKD-EPI 2021	original
`dose_weight_based`	mg/kg arithmetic with a hard maximum given in the input	original
`unit_lab_convert`	mg/dL to mmol/L for common analytes	original
`heart_rate_zones`	Karvonen zones	original
`sleep_cycles`	bed and wake-time planner	original
`nutrition_totals`	totals from a food table given in the input	original

Security (defensive)🔗

Tool	Description	Source
`hash_verify`	MD5, SHA-1, SHA-256, SHA-512, BLAKE2 and verify	stdlib
`password_entropy`	entropy estimate, never echoes the password	original
`sbom_check`	CycloneDX or SPDX JSON required fields and licence summary	original
`jwt_decode`	decode without verifying, with a clear warning	original
`cert_inspect`	PEM certificate fields and expiry (bash with openssl)	original
`secrets_scan`	key-like strings in files	gitleaks ideas
`totp_generate`	RFC 6238 code from a secret in the input	original
`csp_lint`	Content-Security-Policy weaknesses	original

Media🔗

Tool	Description	Source
`image_info`	dimensions and format from PNG, JPEG, GIF headers; EXIF tier 2 with Pillow	Pillow
`qr_generate`	QR code to SVG (tier 2 with `segno`, pure Python)	segno
`wav_info`	duration and format of WAV	stdlib `wave`
`color_tools`	hex, RGB, HSL, WCAG contrast, palette from a seed	original
`srt_tools`	shift, convert, merge subtitles	original
`ascii_banner`	text banner	original

Personal, games, cooking and engineering🔗

Tool	Description	Source
`eisenhower_matrix`	tasks to quadrants with a plan	original
`habit_streaks`	streaks and rates from a log	original
`unit_price_compare`	price per unit across offers	original
`sudoku_solve`, `chess_fen_validate`, `dice_roll_seeded`, `wordle_helper`	puzzles and games	original
`recipe_scale`, `cooking_unit_convert`	scale servings, convert cups and grams by ingredient density	original
`resistor_color`, `ohms_law`, `beam_load`	electronics and statics	original

First batch🔗

Fourteen tier-1 tools, one or two per domain, each the kind of computation a model gets wrong and a person checks: seat_allocation and business_days (done), vote_condorcet, invoice_calc, iban_validate, exact_calc, stats_describe, citation_format, readability, npv_irr, deadline_calc, srs_schedule, tz_convert, hash_verify. Then the Spanish-administration set (es_tax_id_validate, eu_vat_validate, tax_bracket_estimate, transparency_request_draft), which no other catalog serves.

Repositories to learn from🔗

Repository	Stars	Last push	Why
modelcontextprotocol/registry	7.4k	2026-10-14	`server.json`, namespace auth, publisher CLI
anthropics/skills	181.6k	2026-10-09	`SKILL.md` conventions, `marketplace.json`
agentskills/agentskills	26.2k	2026-08-09	spec and `skills-ref` validator
strands-agents/tools	1.3k	2026-10-13	50 tools incl. `calculator`, `current_time`, `rss`, `diagram`
huggingface/smolagents	29.9k	2026-10-06	the smallest clean tool interface
punkpeye/awesome-mcp-servers	96.5k	2026-10-15	category taxonomy
arthurdejong/python-stdnum	602	2026-08-15	IBAN, VAT, NIF/NIE and national ids
vacanza/holidays	1.9k	2026-10-16	holiday data per country and region
hgrecco/pint	2.8k	2026-10-13	units with dimensional analysis
textstat/textstat	1.4k	2026-10-13	readability formulas
numpy/numpy-financial	414	2026-10-05	NPV, IRR, PMT edge cases
collective/icalendar	1.2k	2026-10-15	RFC 5545 reference
wireservice/csvkit, johnkerl/miller	6.4k / 10.0k	2026-09 / 2026-10	CSV verbs
citation-style-language/styles	3.9k	2026-10-13	citation styles
open-spaced-repetition/py-fsrs	513	2026-08-09	spaced repetition
andrewcmyers/civs	96	2026-08-21	Condorcet and Schulze
heuer/segno	814	2026-07-23	pure-Python QR

No maintained small repository exists for D'Hondt and Sainte-Laguë (the best have about 14 stars) or for GDPR templates; those are original tools, which is why seat_allocation is the first reference entry.