# Systemprompt Tools

> Call the free systemprompt.io tools API from any agent: EU AI Act risk classification, CLAUDE.md scoring, llms.txt/AGENTS.md generation and validation, GitHub Actions permissions, and AI ROI calculation. No API key, no signup.

- Skill: `systempromptio/systemprompt-tools` (Agent Skill)
- Install (CLI): `npx skillmds@latest add systempromptio/systemprompt-tools`
- Raw SKILL.md: https://api.skillmd.com/api/skills/systempromptio/systemprompt-tools/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: systempromptio (https://skillmd.com/u/systempromptio)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/systempromptio/systemprompt-tools

---


# systemprompt.io Free Tools

Six agent-callable tools hosted at systemprompt.io. All are deterministic rule engines with instant JSON responses. No authentication beyond a self-chosen agent identifier.

## API contract

- Manifest with full input schemas: `GET https://systemprompt.io/api/v1/tools`
- Every request must carry an `X-Agent-Id` header (or `agent_id` body field): free-form, 1-64 chars. Pick a stable name for your agent and reuse it.
- Rate limit: 60 requests/hour per (IP, agent_id).
- Interactive UIs live at `https://systemprompt.io/tools/{slug}` with a `#reference` section documenting the underlying rules.

## Tools

### EU AI Act risk classifier

Walk the Annex III decision tree. Send `{"answers": []}` to get the first question; append one 0-based option index per answer and resend until you receive a classification with risk class, cited articles, obligations, and an evidence checklist.

```bash
curl -s -X POST https://systemprompt.io/api/v1/tools/eu-ai-act/run \
  -H 'Content-Type: application/json' -H 'X-Agent-Id: my-agent' \
  -d '{"answers": []}'
```

### CLAUDE.md scorer

```bash
curl -s -X POST https://systemprompt.io/api/v1/tools/claude-md-scorer/run \
  -H 'Content-Type: application/json' -H 'X-Agent-Id: my-agent' \
  -d "$(jq -n --rawfile c CLAUDE.md '{content: $c}')"
```

Returns a score out of 100, grade, and per-rule pass/fail with recommendations. Useful as a CI gate for CLAUDE.md quality.

### llms.txt / AGENTS.md generator and validator

Validate: `{"mode": "validate", "content": "<file text>", "target": "llms"}` (or `"agents"`).
Generate: `{"mode": "generate", "fields": {"site_name": ..., "description": ..., "sections": [{"title": ..., "links": [{"label": ..., "url": ...}]}]}}`.
Endpoint: `POST https://systemprompt.io/api/v1/tools/llms-txt/run`.

### GitHub Actions permissions generator

`{"tasks": ["checkout", "push-commits", ...]}` to `POST .../tools/github-actions-permissions/run` returns a minimal `GITHUB_TOKEN` permissions block with fork and GitHub App caveats. Valid task ids are listed in the manifest.

### AI ROI calculator

`{"team_size", "seat_cost_monthly", "adoption_pct", "hours_saved_per_week", "loaded_hourly_cost", "implementation_cost?"}` to `POST .../tools/ai-roi-calculator/run` returns monthly cost, value, net ROI, break-even months, and a 0.5x-1.5x sensitivity range.

## AI deep review (email-gated)

The CLAUDE.md scorer and llms.txt tools also offer an LLM deep review at `POST .../tools/{tool}/deep-review`. These require an `email` field in the body and are limited to 5 requests/hour per (IP, agent_id). Only use with an email the user has provided for this purpose.

## MCP alternative

Prefer MCP? The same tools are served over anonymous streamable HTTP by systemprompt.io itself (no OAuth, no key):

```bash
claude mcp add --transport http systemprompt-tools https://systemprompt.io/api/v1/mcp/systemprompt-tools/mcp
```

MCP tool names: `classify_eu_ai_act_risk`, `score_claude_md`, `validate_llms_txt`, `generate_llms_txt`, `generate_github_actions_permissions`, `calculate_ai_roi`.

## Workflow ideas

- Classify a product under the EU AI Act during design review and attach the evidence checklist to the ticket (high-risk obligations enforce from 2026-08-02).
- Score CLAUDE.md in CI on every PR that touches it; fail below a threshold.
- Generate llms.txt and AGENTS.md for a site you are building, then validate on deploy.
- Compute minimal workflow permissions when authoring GitHub Actions instead of defaulting to `write-all`.