Skip to content

andoop/agentx-platform

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

AgentX

English | 简体中文

Status: Alpha License: Apache--2.0 Docs: EN/ZH npm GitHub Release

Quick Links: Quick Install | Quick Start | 5-Minute Trial | CLI Installation | Docs Index | Changelog | Release Guide | Contributing | Security

AgentX is an open-source registry platform for managing and distributing AI agent artifacts across teams.

It gives you one place to publish, review, approve, version, and install:

  • MCP Server
  • Skill
  • Command

AgentX ships with a Web console, Registry API, packaged CLI tooling, and local adapters for Cursor and Claude Code.

Quick Install

Install the published CLI from npm:

npm install -g agentc
agentc --help

Compatibility alias:

agentx --help

Quick Start

  1. Install the CLI with npm install -g agentc
  2. Start the local stack with Docker Compose or run the API and Web app from source
  3. Sign in as alice@internal
  4. Publish a skill, command, or mcp_server
  5. Review and approve it
  6. Install it into your local agent runtime with the CLI

Status

AgentX is currently alpha.

What works today:

  • unified artifact model for mcp_server, skill, and command
  • Web catalog, publish, review, and settings pages
  • JWT auth with role-based access control
  • file persistence by default, Postgres via Prisma optionally
  • package bundle upload and distribution
  • install adapters for Cursor and Claude Code

What is still evolving:

  • the first public npm release process for the CLI
  • Docker-based quick deployment
  • broader automated test coverage
  • richer search, governance, and package lifecycle workflows

Why AgentX

Teams adopting AI tools usually end up with fragmented assets:

  • MCP servers live in one repo
  • prompts and skills live in another
  • slash commands are copied manually between projects
  • approval, ownership, and audit trails are inconsistent

AgentX provides a single workflow for:

  1. authoring an artifact
  2. publishing it to a registry
  3. reviewing and approving it
  4. distributing it to local agent runtimes

Core Features

  • Unified schema for MCP Server, Skill, and Command
  • Web console for catalog browsing, publishing, review, and admin operations
  • CLI workflows for login, publish, review, install, export, import, and backup
  • Package-aware distribution: publish a manifest plus real artifact files together
  • Local install targets for Cursor and Claude Code
  • Audit logs, member management, and role-based governance
  • File-based persistence by default for low-friction local development

5-Minute Trial

The fastest way to try AgentX is Docker Compose:

cp .env.docker.example .env
docker compose up --build

Then open:

  • http://localhost:3000/login

The stack includes:

  • web on port 3000
  • api on port 4000
  • postgres on port 5432

If those ports are already in use, change them in .env before starting Compose:

AGENTX_WEB_PORT=3003
AGENTX_API_PORT=4010
POSTGRES_PORT=5433

If you prefer running from source, install dependencies and start the API and Web app:

npm install
npm run dev:api
npm run dev:web

Use the default demo credentials:

  • alice@internal / agentx123
  • bob@internal / agentx123

Then try this flow:

  1. Sign in as alice@internal
  2. Open /publish
  3. Create a command or skill
  4. Open /review
  5. Approve the submitted review request
  6. Install the artifact into a local project with the CLI

CLI Installation

The CLI is available on npm:

npm install -g agentc
agentc --help

The legacy agentx command remains available as a compatibility alias:

agentx --help

If you want to test a local build before publishing, you can still build and pack it manually:

npm install
npm run pack:cli

Then install the generated tarball globally:

npm install -g ./apps/cli/agentc-0.1.1.tgz
agentc --help

For local development, you can still run it from source:

npm --workspace agentc run dev -- login --api-url http://localhost:4000 --email alice@internal --password agentx123

npm --workspace agentc run dev -- publish artifact.example.yaml
npm --workspace agentc run dev -- publish artifact.skill.example.yaml
npm --workspace agentc run dev -- publish artifact.command.example.yaml

npm --workspace agentc run dev -- reviews
npm --workspace agentc run dev -- review <review_id> --decision approved --notes "looks good"

npm --workspace agentc run dev -- install github-mcp --agent cursor --dir /path/to/your/project
npm --workspace agentc run dev -- install review-playbook --agent cursor --dir /path/to/your/project
npm --workspace agentc run dev -- install release-summary --agent cursor --dir /path/to/your/project

Distribution Model

AgentX supports both manifest-only artifacts and package-backed artifacts.

When an artifact includes a package bundle:

  • the package is uploaded together with the manifest
  • AgentX stores the package per slug + version
  • install and sync extract the package into .agentx/packages/<slug>/<version>/
  • local adapters then generate the target agent files

Current install targets:

  • mcp_server
    • Cursor: .cursor/mcp.json
    • Claude Code: .mcp.json
  • skill
    • Cursor: .cursor/rules/<slug>.mdc
    • Claude Code: .claude/skills/<slug>/SKILL.md
  • command
    • Cursor: .cursor/commands/<slug>.md
    • Claude Code: .claude/commands/<slug>.md

Package conventions in V1:

  • skill packages may include cursor/<slug>.mdc and claude/SKILL.md
  • command packages may include cursor/<slug>.md and claude/<slug>.md
  • mcp_server packages may include the real script or binary referenced by spec.entrypoint

Persistence

By default, the API uses file persistence:

  • data file: apps/api/.data/agentx.json
  • backup directory: apps/api/.data/backups

This means local development works without any external database.

To override the file location:

AGENTX_DATA_FILE=/absolute/path/agentx.json npm run dev:api

To use Postgres with Prisma:

cp apps/api/.env.example apps/api/.env
npm --workspace @agentx/api run prisma:generate
npm --workspace @agentx/api run db:push
npm run dev:api

The health endpoint reports the active repository mode:

  • GET /health

Docker Demo

AgentX includes a self-hosted demo stack for local evaluation.

Quick start:

cp .env.docker.example .env
docker compose up --build

Or use the npm helpers:

npm run docker:up
npm run docker:down

Default services:

  • Web: http://localhost:3000
  • API: http://localhost:4000
  • Postgres: localhost:5432

The Compose stack automatically:

  • starts Postgres
  • waits for database readiness
  • runs prisma db push before starting the API
  • starts the Web app after the API becomes healthy

Auth And Roles

AgentX uses JWT-based auth.

  • API clients use Bearer tokens
  • the Web app stores auth in an HttpOnly cookie
  • the CLI stores session data in ~/.agentx/session.json

Default roles:

  • viewer
  • publisher
  • reviewer
  • admin

Admin capabilities currently include:

  • create members
  • update teams and roles
  • reset passwords
  • disable and enable members
  • prevent removal of the last active admin

Web Console

Current pages:

  • /login
  • /catalog
  • /publish
  • /review
  • /settings

The Web publish page supports:

  • type-specific artifact fields
  • package directory upload
  • inline form validation and status feedback

Workspace Layout

  • apps/web: Next.js management console
  • apps/api: Fastify Registry API
  • apps/cli: Node.js CLI
  • packages/schema: shared domain schema and validation
  • packages/config: artifact.yaml parsing and templating
  • packages/sdk: API client SDK
  • packages/ui: shared Web UI components
  • docs/architecture: architecture notes

Development

Common commands:

npm run typecheck
npm run build
npm run test
npm run test:api
npm run test:cli
npm run test:web
npm run dev:api
npm run dev:web
npm run dev:cli

Persistence admin commands:

npm --workspace agentc run dev -- export --out ./agentx-export.json
npm --workspace agentc run dev -- import ./agentx-export.json
npm --workspace agentc run dev -- backup --label before-migration

Architecture

Start with:

  • docs/en/architecture/overview.md
  • docs/en/architecture/mvp.md

Runtime flow:

  1. publish an artifact from Web or CLI
  2. create a review request
  3. approve or reject it through governance workflows
  4. store artifact state through a unified repository layer
  5. install the approved artifact into local agent runtimes

Limitations

  • CLI packaging is ready, but the first public npm release process is still pending
  • smoke coverage exists, but deeper contract, failure-mode, and adapter compatibility tests are still limited
  • package upload in the Web UI depends on browser directory-upload support
  • the project is still early and APIs may evolve

Roadmap

  • publish the packaged CLI as a public npm release
  • slim and harden Docker images for production deployment
  • improve search and artifact discovery
  • add stronger package lifecycle rules and provenance metadata
  • expand automated API, CLI, and Web test coverage beyond smoke paths
  • improve multi-language documentation

Contributing

Please read:

  • CONTRIBUTING.md / CONTRIBUTING.zh-CN.md
  • CODE_OF_CONDUCT.md / CODE_OF_CONDUCT.zh-CN.md
  • SECURITY.md / SECURITY.zh-CN.md

Contributions are welcome for:

  • docs
  • adapters
  • governance workflows
  • artifact package lifecycle
  • testing and CI

License

Apache-2.0. See LICENSE.

About

Open-source AI agent artifact registry for MCP servers, skills, and commands.

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors