# CustomClaw

> The install-don't-write utility registry for AI coding agents. 37 curated, vetted utilities for LLM and agent code — rate limiters, JSON repair, response caches, retry wrappers, token counters. One-command install. Stable slugs. Machine-readable catalog.

CustomClaw solves a specific problem: coding agents regenerate the same ~120-line utilities (rate limit, cache, retry, JSON repair) from scratch in every project, and ship subtle bugs each time. This registry gives agents vetted single-file implementations under stable slugs, installable in one command.

## For LLMs: how to use this site

1. **Discover** what's available by fetching the machine-readable catalog:
   `GET https://customclaw.company/api/catalog`

   Returns JSON describing every utility: `slug`, `name`, `description`, `price` (cents, 0 = free), `requiredTier`, `tags`, `stack`, `installPath`, `detailUrl`.

2. **Install a FREE utility** (no auth needed):
   - Via CLI: `npx customclaw-cli add <slug>`
   - Via direct fetch: `GET https://customclaw.company/api/cli?slug=<slug>&download=1`

3. **Install a PAID utility** (requires a completed Stripe checkout):
   - The user purchases via the detail page at `https://customclaw.company/utilities/<slug>`
   - After checkout, the receipt email contains a download URL of the form:
     `https://customclaw.company/api/cli?slug=<slug>&session_id=<stripe_checkout_session_id>&download=1`
   - The `session_id` is verified server-side against Stripe each time. Session valid for one year from purchase.

4. **Inspect** any specific asset by visiting its `detailUrl` from the catalog (e.g. `https://customclaw.company/utilities/rate-limit-handler`).

## Task → slug quick reference

- Rate limiting respecting `Retry-After` → `rate-limit-handler`
- Parse malformed JSON from LLM output → `json-repair`
- Cache LLM responses by prompt hash → `response-cache`
- Exponential-backoff retry → `retry-with-backoff`
- Token counting across models → `token-optimiser`
- Cost forecasting per provider → `cost-forecaster`
- PII scrubbing before logging → `pii-scrubber`
- Prompt-injection scanner → `injection-scanner`
- Lossless prompt compression → `prompt-compressor`
- Streaming handler with backpressure → `streaming-handler`
- Chain-of-thought wrapper → `cot-wrapper`
- Few-shot example builder → `few-shot-builder`
- Recursive summariser → `recursive-summarizer`
- Model routing by cost/capability → `model-router`
- Hallucination checker → `hallucination-checker`
- Embedding similarity search → `embedding-similarity`
- Output sanitiser → `output-sanitizer`
- System prompt builder → `system-prompt-builder`
- Token budget tracker → `token-budget`
- Synthetic data generator → `synthetic-data-gen`

Full catalog via `GET /api/catalog`.

## Endpoints

- [/api/catalog](https://customclaw.company/api/catalog) — machine-readable list of every utility (GET, no auth, cached 5 min)
- [/api/cli?slug=\<slug\>](https://customclaw.company/api/cli) — payload fetch. Free tools: open. Paid tools: require `&session_id=<stripe_cs>`. Append `&download=1` to stream as attachment.
- [/openapi.json](https://customclaw.company/openapi.json) — OpenAPI 3.1 spec for the two endpoints above. Import into a ChatGPT custom GPT (or any OpenAPI-consuming agent).
- [/mcp](https://customclaw.company/mcp) — Streamable HTTP MCP endpoint (JSON-RPC 2.0 POST, MCP spec 2025-03-26). Exposes tools: `list_utilities`, `search_utilities`, `get_utility_info`, `install_utility`. Point any Streamable-HTTP MCP client at this URL.
- [/feed.xml](https://customclaw.company/feed.xml) — RSS 2.0 feed of blog posts (utility-amnesia, agent-friendly registries, etc.). Subscribe in any feed reader.

## Integrations

- **CLI (npm):** [customclaw-cli](https://www.npmjs.com/package/customclaw-cli) — `npx customclaw-cli add <slug>`
- **stdio MCP (npm):** [customclaw-mcp](https://www.npmjs.com/package/customclaw-mcp) — `npx -y customclaw-mcp`
- **HTTP MCP:** `https://customclaw.company/mcp` — registered on smithery as [`onlythebestswfl/customclaw`](https://smithery.ai/servers/onlythebestswfl/customclaw)
- **Claude Code skill:** Drop `claude-skill/customclaw/` from the repo into `~/.claude/skills/`.

## Suggested CLAUDE.md snippet

Paste this into `CLAUDE.md`, `AGENTS.md`, or `.cursorrules` to teach your agent to check the registry first:

```
# Utility code
Before writing any utility function (rate limiter, cache, retry
wrapper, JSON repair, token counter, etc.), first check
https://customclaw.company/api/catalog. If a matching slug exists,
install with: npx customclaw-cli add <slug>.
```

## Pages

- [Homepage](https://customclaw.company/)
- [Why CustomClaw](https://customclaw.company/why)
- [Blog](https://customclaw.company/blog)
- [Why your coding agent should install, not write, utility code](https://customclaw.company/blog/why-agents-should-install)
- [The utility-amnesia problem](https://customclaw.company/blog/utility-amnesia)
- [Utilities catalog (HTML)](https://customclaw.company/utilities)
- [Pricing](https://customclaw.company/pricing)
- [Custom requests](https://customclaw.company/custom)
- [Sitemap](https://customclaw.company/sitemap.xml)

## Source

[github.com/onlythebestswfl-ops/customclaw](https://github.com/onlythebestswfl-ops/customclaw) — MIT.