StackSherpa

stacksherpa

Intelligent API recommendation engine. 450+ providers across 28 categories — with pricing, known issues, compliance data, and live GitHub issue tracking.

Browse the catalog →

Quick start

# 1. Install the CLI
npm install -g stacksherpa

# 2. Install the Claude Code skill (Claude will auto-consult stacksherpa when you need an API)
mkdir -p ~/.claude/skills/api-selection
curl -sL https://raw.githubusercontent.com/astein91/stacksherpa/main/skills/api-selection/SKILL.md \
  -o ~/.claude/skills/api-selection/SKILL.md

# 3. Try it
stacksherpa providers email --pretty

No API keys needed. The catalog is a shared read-only database.

---

Browse the catalog

Visit stacksherpa.vercel.app to explore providers, compare options, and see pricing and known issues — no install required.

CLI usage

stacksherpa categories --pretty                # List all 28 categories
stacksherpa providers email --pretty            # All email providers + your profile
stacksherpa providers database --pretty         # All database providers
stacksherpa provider stripe --pretty            # Deep-dive on one provider

stacksherpa profile --pretty                    # View your project profile
stacksherpa profile --set '{"project.scale": "startup"}'
stacksherpa profile --set '{"constraints.compliance": ["SOC2"]}'

stacksherpa decide --api Resend --category email --outcome positive
stacksherpa report --id <decision-id> --success

stacksherpa projects list --pretty

All commands output JSON. Add --pretty for human-readable formatting.

Claude Code skill

The quick start installs the skill globally (~/.claude/skills/). You can also install per-project:

mkdir -p .claude/skills/api-selection
curl -sL https://raw.githubusercontent.com/astein91/stacksherpa/main/skills/api-selection/SKILL.md \
  -o .claude/skills/api-selection/SKILL.md

Once installed, Claude will silently run stacksherpa providers <category> before suggesting any API integration. No slash command needed — the skill triggers automatically.

Alternative — no skill, just instructions: Add this to your project's CLAUDE.md:

## API Selection
When implementing features that require external APIs:
1. Run `stacksherpa providers <category>` via Bash before choosing a provider
2. Pick the best match based on the returned profile, pricing, and issues
3. After integration, run `stacksherpa decide --api X --category Y --outcome positive`

Categories (28)

Group	Categories
Core	auth, database, storage, email, payments, analytics, monitoring, search
Messaging	sms, push, realtime, chat
AI	ai, ai-audio, ai-video, ai-image, vector-db
Infrastructure	cache, jobs, message-queue, hosting, cdn, feature-flags
Content & data	cms, media, maps, web-search, finance

What each provider includes

• Description, website, docs URL, pricing URL
• Strengths (dx, reliability, cost, performance, support, security, customization)
• Best for (hobby / startup / growth / enterprise)
• Avoid if / best when (concrete conditions)
• Alternatives (linked provider IDs)
• Compliance certs (SOC2, HIPAA, GDPR, PCI-DSS, ISO27001)
• Ecosystem affinity (vercel, supabase, aws, cloudflare, etc.)
• Self-hostable flag
• Live pricing — scraped from provider websites
• Known issues — scraped from GitHub (reactions, severity, workarounds)

Project profile

Create .stacksherpa/profile.json in your project root to personalize recommendations:

{
  "project": {
    "name": "my-app",
    "stack": { "language": "TypeScript", "framework": "Next.js", "hosting": "Vercel" },
    "scale": "startup"
  },
  "constraints": {
    "compliance": ["SOC2"],
    "budgetCeiling": { "monthly": 100 }
  },
  "preferences": {
    "prioritize": ["dx", "reliability"],
    "avoidProviders": ["sendgrid"]
  }
}

Global defaults in ~/.stacksherpa/defaults.json apply to all projects (local profiles override).

Data architecture

Turso (shared, read-only for CLI clients)
  providers      — 450+ providers with rich metadata
  pricing        — versioned pricing snapshots
  known_issues   — scraped from GitHub (reactions, severity)
  discovery_log  — audit trail of bootstrap runs

Local (~/.stacksherpa/)
  defaults.json  — global defaults across all projects

Local (<project>/.stacksherpa/)
  profile.json   — project constraints & preferences
  decisions.json — API selection history

Your data never leaves your machine. The Turso catalog is read-only.

---

Keeping the catalog fresh

The catalog is maintained by an agentic pipeline — no manual data entry.

Step	Schedule	What it does
GitHub issues	Every run	Scrapes issues with 2+ reactions from provider repos
Pricing	Every run	Re-scrapes up to 3 stale providers via Firecrawl
Discovery	1st & 15th	Finds new providers via Exa search
Agent refresh	Daily (AI), Mondays (all)	Claude Haiku re-evaluates provider profiles with live web data
Bootstrap	1st of month	Full agentic discovery across all 28 categories

Provider review

npm run cron:review -- --list                    # See pending providers
npm run cron:review -- --approve pinecone qdrant # Approve specific ones
npm run cron:review -- --reject some-blog        # Reject bad entries

Environment variables (for maintainers)

Variable	Required for	Purpose
TURSO_AUTH_TOKEN	All reads	Read-only Turso access
TURSO_WRITE_TOKEN	Cron jobs	Write access for scrapers/agents
ANTHROPIC_API_KEY	Agent refresh, bootstrap	Claude Haiku for agentic pipelines
EXA_API_KEY	Bootstrap	Web search for provider discovery
FIRECRAWL_API_KEY	Pricing, metadata	Website scraping
GITHUB_TOKEN	Issue scraping	Higher rate limits for GitHub API

Development

npm install
npm run build                                     # Type-check
npm run dev -- providers email --pretty            # Run CLI locally
npm test                                           # Run tests
npm run cron:bootstrap -- --category jobs --dry-run # Bootstrap one category
npm run cron:agent-refresh -- --full               # Full agent refresh

License

MIT