Skip to content

Benja-Pauls/SerpentStack

SerpentStack

SerpentStack

npm Release CI MIT License

Find any AI agent skill or MCP server. Search every registry, install in seconds.

serpentstack search demo

Quick Start · Search · Install · Discover · Base Skills · Persistent Agents


Agent skills are how Stripe merges 1,300+ agent-generated PRs per week and Shopify operates with agents as a core part of engineering. But skills are scattered across registries with no unified way to find or install them.

SerpentStack searches every major skill registry and MCP server directory at onceAnthropic, skills.sh, awesome-agent-skills, public GitHub, and the MCP Registry — with a single command. One CLI to search, install, and manage skills and discover MCP servers for any AI coding agent.

npm install -g serpentstack

Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, Windsurf, and any tool that reads the Agent Skills open standard. Zero dependencies. Requires Node 22+.


Why SerpentStack?

Skills and MCP servers are scattered across isolated registries. There's no unified way to find what exists.

skills.sh Anthropic awesome-agent-skills Glama mcp.so SerpentStack
Skills all
MCP servers all
Single CLI search
Install from any source
Project-aware recommendations
Auto-generate missing skills

One search, every source. Skills and MCP servers together, clearly labeled.


Quick Start

serpentstack search "react testing"        # search every registry
serpentstack add clerk                     # install by name
serpentstack add stripe/agent-toolkit      # install from a GitHub repo
serpentstack discover                      # get recommendations for your project

That's it. Skills are installed to .skills/<name>/SKILL.md and your agents pick them up automatically.


Search

Search across all major skill registries and MCP server directories with one command. Results are ranked by relevance, clearly split into Skills and MCP Servers so you always know what you're looking at.

serpentstack search "auth oauth"
serpentstack search "stripe payments"
serpentstack search "postgres database"
serpentstack search "docker deploy"
  ── Skills (3) ──────────────────────────────────────

   1. clerk  anthropic
      Authentication and user management with Clerk
      $ serpentstack add anthropics/skills/clerk

   2. better-auth  skills.sh
      Type-safe authentication framework for TypeScript
      $ serpentstack add better-auth/best-practices

   3. oauth-patterns  awesome
      OAuth 2.0 and OpenID Connect integration patterns
      $ serpentstack add nichochar/oauth-patterns

  ── MCP Servers (2) ─────────────────────────────────

   1. supabase  mcp
      Manage Supabase projects, databases, edge functions, and auth
      Claude Code: claude mcp add supabase -- npx -y supabase-mcp-server
      Docs:        github.com/supabase-community/supabase-mcp

   2. neon  mcp
      Serverless Postgres with branching, schema migrations, and database management
      Claude Code: claude mcp add --transport http neon https://mcp.neon.tech/sse
      Docs:        github.com/neondatabase/mcp-server-neon

Skills and MCP servers are clearly labeled: anthropic skills.sh awesome github for skills, mcp for MCP servers. Skills teach agents how to build; MCP servers give agents tools to use — you often want both.


Install

Install skills from any source — a registry name, a GitHub repo, or a URL. SerpentStack resolves the source, fetches the SKILL.md, and drops it into your .skills/ directory.

serpentstack add clerk                           # resolve via registries
serpentstack add anthropics/skills/clerk          # specific GitHub path
serpentstack add stripe/agent-toolkit             # GitHub repo
serpentstack add --force clerk                    # overwrite existing

If no community skill exists, SerpentStack auto-generates a context-rich stub by pulling the project's README, package metadata, install commands, and docs URL from GitHub. You get a useful starting point instead of nothing.

serpentstack add fastify
# → Generated fastify → .skills/fastify/SKILL.md
#   Source: github.com/fastify/fastify
#   Docs:   fastify.dev

Discover

Analyze your project and get skill and MCP server recommendations tailored to your actual stack. SerpentStack reads your dependency files, detects frameworks/services/tools, and searches registries for matching skills and MCP servers — skipping skills you already have installed.

cd your-project
serpentstack discover
  Your stack
  Languages:    javascript, typescript
  Frameworks:   react, nextjs
  Services:     clerk, stripe, prisma
  Tools:        vitest, tailwindcss, eslint

  ── Services ────────────────────────────────────────
  clerk  anthropic
  $ serpentstack add anthropics/skills/clerk

  stripe-best-practices  skills.sh
  $ serpentstack add stripe/agent-toolkit
  ...

  ── MCP Servers (3) ─────────────────────────────────
  stripe  mcp
  $ claude mcp add stripe -- npx -y @stripe/mcp

  neon  mcp
  $ claude mcp add --transport http neon https://mcp.neon.tech/sse
  ...

  12 skill recommendations, 3 MCP servers based on your stack.
  3 skills already installed.

Detects: JavaScript/TypeScript, Python, Go, Rust, Ruby — and 50+ frameworks, services, and tools within each.


Base Skills

SerpentStack ships 10 production-quality skills you can install into any project. These aren't descriptions — they contain complete, copy-paste templates with real imports and type signatures. Every skill ends with a verification step so agents can confirm their own work.

serpentstack skills                     # install all 10 base skills
serpentstack skills update              # update to latest versions
Skill What it teaches agents
scaffold End-to-end resource generation: model, schema, service, route, migration, tests, frontend types
auth JWT, ownership enforcement, SSO swap patterns (one function to replace)
test Real Postgres via testcontainers, savepoint isolation, async httpx
db-migrate Alembic workflow: create, review, apply, troubleshoot, rollback
deploy Docker multi-stage builds, ECR push, Terraform plan/apply, rollback
dev-server Error detection patterns for FastAPI and Vite/React
git-workflow Branch naming, conventional commits, PR checklists
model-routing Delegate code generation to local models for 10-50x cost reduction
generate-skills Interview-based generation of project-specific skills for any codebase
find-skills Evaluate and adopt community skills safely

The most powerful is generate-skills — ask your agent to read .skills/generate-skills/SKILL.md and it will analyze your codebase, interview you about your decisions, and produce a custom skill set for your project. This is the fastest path to agents that understand your conventions.


Persistent Agents

SerpentStack also includes three background agents that run on local models via Ollama at zero cost. They watch your project continuously — catching crashes, running tests, and keeping your skills accurate as code evolves.

Setup and usage
serpentstack persistent                  # guided setup on first run
serpentstack persistent --start          # launch agents
serpentstack persistent --stop           # stop all agents
serpentstack persistent --agents         # change models or enable/disable
serpentstack notifications               # see what your agents found
.openclaw/
  SOUL.md                     # shared context inherited by all agents
  agents/
    log-watcher/AGENT.md      # dev server health, crash detection     (every 30-60s)
    test-runner/AGENT.md       # test suite, lint, typecheck            (every 5-15min)
    skill-maintainer/AGENT.md  # skill drift detection                  (every 1hr)

Log Watcher monitors dev server output and catches crashes, import failures, and runtime errors with file paths and fix suggestions.

Test Runner runs your test suite, linter, and type checker on a schedule. Tracks which tests are failing, what changed, and whether the test or the source needs updating.

Skill Maintainer compares .skills/ files against actual code patterns and proposes updates when conventions drift. Without this, skills go stale and agents start producing code that doesn't match your project.

During setup, SerpentStack installs Ollama and downloads a model if needed — no API keys required. To add your own agent, create a folder under .openclaw/agents/ with an AGENT.md file.


Production Template

Projects created with serpentstack stack new include a complete fullstack application with all skills and agents pre-configured. This is not a starter template with TODO comments — it ships with working authentication, resource CRUD, ownership enforcement, database migrations, and infrastructure-as-code.

Architecture and conventions

Prerequisites: Python 3.12+, Node 22+, Docker, uv

serpentstack stack new my-app
cd my-app
make init && make setup && make dev
backend/        FastAPI + async SQLAlchemy + asyncpg + Alembic
frontend/       React + TypeScript + Vite + Tailwind + React Query
infra/          Terraform: App Runner, RDS, ECR, VPC (dev/staging/prod)
.skills/        All 10 base skills pre-installed
.openclaw/      Persistent agent workspace pre-configured

Key conventions (encoded in .skills/ so agents learn them automatically):

  • Services flush, routes commit. Services call db.flush() but never db.commit(). Routes own the transaction boundary.
  • Services return domain objects, not HTTP errors. None = not found, False = forbidden, domain object = success. Routes translate to HTTP status codes.
  • Auth is one function. All protected routes depend on get_current_user(). Swapping JWT for Clerk/Auth0 means replacing one dependency.
  • Types flow from backend to frontend. make types generates TypeScript interfaces from the OpenAPI spec. No hand-written API types.
make dev             # Postgres + Redis + backend + frontend with hot reload
make verify          # lint + typecheck + test (both stacks)
make deploy          # build, push, terraform apply

CLI Reference

# Search & install
serpentstack search <query>             # search skills + MCP servers across all registries
serpentstack add <source>               # install a skill from any registry or repo
serpentstack discover                   # analyze project, recommend skills + MCP servers

# Base skills
serpentstack skills                     # download base skills + agent configs
serpentstack skills update              # update to latest versions

# Persistent agents
serpentstack persistent                 # status dashboard (guided setup on first run)
serpentstack persistent --start         # launch agents
serpentstack persistent --stop          # stop all agents
serpentstack notifications              # what your agents found

# Template
serpentstack stack new <name>           # scaffold a new project
serpentstack stack update               # update template files

Contributing

Contributions are welcome — see CONTRIBUTING for setup and guidelines. The highest-impact areas:

  • MCP server entries — add popular servers to the curated registry in registries.js
  • Registry adapters — connect additional skill or MCP sources
  • Stack detection — detect more languages, frameworks, and tools in discover
  • New skills for popular frameworks and services (follow SKILL-AUTHORING.md)

Open an issue for bugs and feature requests.

License

MIT