npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

cap-protocol

v0.1.1

Published

A capability registry layer for AI agents — versioned, lifecycle-aware, auditable.

Readme

cap-protocol

The capability registry layer between MCP and the agent's planner.

npm version License: MIT CI

What it is

A protocol and reference implementation for registering, versioning, and evolving the capabilities an AI agent can use — tools, prompts, sub-agents, memory stores, integrations.

Today, agent systems built on MCP and A2A turn into spaghetti: flat lists of tools with no version, no lifecycle, no audit trail. When something breaks you can't tell when it broke, why, or how to roll back. cap-protocol gives those resources a state machine and an append-only event log.

What it solves

| Today | With cap-protocol | |---|---| | status: active is the only field | state is a finite-state machine: proposed → registered → verified → active → degraded → deprecated → archived | | No version per capability | Every resource has a semver version and a versioned interface | | Capability list regenerated wholesale | propose → assess → commit → rollback per change, append-only event log | | "Was this tool working last Tuesday?" → git blame | One CLI query: cap history <id> --at 2026-04-22 |

30-second example

npm install -g cap-protocol

# Initialize a registry
cap init my-agent-registry

# Propose a new capability
cap propose --root my-agent-registry \
  --id tool_read \
  --layer cc-native \
  --source Read \
  --what "Read files from local filesystem"
# → prints a run-id, e.g. 2026-04-30T07-42-00-000Z_a1b2c3d4

# Validate the proposal (no side effects)
cap assess --root my-agent-registry <run-id>

# Commit it (writes resource + event log entry)
cap commit --root my-agent-registry <run-id>

# Inspect
cap show --root my-agent-registry tool_read
cap history --root my-agent-registry tool_read

# Verify the whole registry against schema + secret/PII patterns
cap verify --root my-agent-registry

# Roll back if needed
cap rollback --root my-agent-registry <event-id>

How it relates to MCP / A2A / AGP

  • MCP says "here are the tools you can call right now." Anthropic owns this.
  • A2A says "here is how agents talk." Google owns this.
  • AGP (Zhang et al. 2026) says "here is how agents evolve themselves." Academic, no widely-adopted implementation.
  • cap-protocol says "here is how an agent knows what it can do and proves it" — the registry layer between MCP and the agent's planner.

cap-protocol is a practitioner subset of AGP: it adopts AGP's two-layer split (resource substrate + closed-loop evolution) but drops the RL-style optimizer layer. It's the bookkeeping discipline you can drop into any agent system today.

See docs/comparison-with-agp.md for the formal mapping.

Architecture

┌──────────────────────────────────────────────────┐
│  Agent Planner ("which capability do I use?")    │
│                       │                          │
│                       ▼                          │
│  cap-protocol Registry                           │
│  ├── resources/<id>.yaml  ← versioned, stateful  │
│  ├── events/YYYY-MM/...   ← append-only log      │
│  └── manifests/<layer>.yaml                      │
│                       │                          │
│                       ▼                          │
│  MCP / Native Tools / Sub-Agents                 │
└──────────────────────────────────────────────────┘

Documentation

Status

v0.1 — protocol spec frozen, TypeScript reference implementation, public example registry. Production agents experimenting in private registries. API surface stable for the v0.x line; semver discipline starts at v1.0.

Contributing

See CONTRIBUTING.md. Issues and PRs welcome — especially around schema edge cases, additional example registries from real agent systems, and adapter packages for Python / Go.

Citation

If cap-protocol helps your research or product, please cite:

@software{webspot2026capprotocol,
  title  = {cap-protocol: A Capability Registry Layer for AI Agents},
  author = {Webspot},
  year   = {2026},
  url    = {https://github.com/webspotme/cap-protocol}
}

And cite the AGP paper that inspired the two-layer architecture:

@misc{zhang2026autogenesisselfevolvingagentprotocol,
  title         = {Autogenesis: A Self-Evolving Agent Protocol},
  author        = {Wentao Zhang and Zhe Zhao and Haibin Wen and Yingcheng Wu and
                   Ming Yin and Bo An and Mengdi Wang},
  year          = {2026},
  eprint        = {2604.15034},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI}
}

License

MIT