CAAMP - Central AI Agent Managed Packages

One CLI to manage Skills, MCP servers, and instruction files across 44 AI coding agents.

CAAMP is a unified provider registry and package manager for AI coding agents. It replaces the need to manually configure each agent's MCP servers, skills, and instruction files individually -- handling the differences in config formats (JSON, JSONC, YAML, TOML), config keys (mcpServers, mcp_servers, extensions, mcp, servers, context_servers), and file paths across all supported providers.

CAAMP adopts LAFS for agent-facing output contracts. Protocol authority lives in the standalone LAFS repository and package.

Install

# Global install (recommended)
npm install -g @cleocode/caamp

# Or run directly with npx
npx @cleocode/caamp <command>

After global install, use caamp directly:

caamp providers list
caamp providers detect
caamp mcp install @anthropic/mcp-server-fetch
caamp skills install owner/repo

Library Usage

CAAMP also exports a full programmatic API:

npm install @cleocode/caamp

import {
  getAllProviders,
  getInstalledProviders,
  listAllMcpServers,
  installSkill,
  detectAllProviders,
} from "@cleocode/caamp";

// Get all registered providers
const providers = getAllProviders();

// Detect which agents are installed on this system
const installed = getInstalledProviders();

// List MCP servers across all installed providers
const servers = await listAllMcpServers(installed, "global");

See API Reference for full programmatic API documentation.

CLI Commands

Providers

caamp providers list               # List all supported providers
caamp providers list --tier high   # Filter by priority tier
caamp providers detect             # Auto-detect installed providers
caamp providers detect --project   # Detect project-level configs
caamp providers show <id>          # Show provider details + all paths

Skills

caamp skills install <source>      # Install from GitHub/URL/marketplace
caamp skills remove [name]         # Remove skill(s) + symlinks
caamp skills list [-g]             # List installed skills
caamp skills find [query]          # Search marketplace (agentskills.in + skills.sh)
caamp skills find [query] --recommend --top 5   # Ranked recommendations + CHOOSE flow
caamp skills find [query] --recommend --json    # LAFS envelope JSON for agent consumers
caamp skills init [name]           # Create new SKILL.md template
caamp skills validate [path]       # Validate SKILL.md format
caamp skills audit [path]          # Security scan (46+ rules, SARIF output)
caamp skills check                 # Check for available updates
caamp skills update                # Update all outdated skills

Recommendation criteria flags:

--must-have <term>   # repeatable and comma-delimited
--prefer <term>      # soft preference signal
--exclude <term>     # hard exclusion signal
--details            # expanded JSON evidence fields
--select <index>     # select from ranked CHOOSE list

LAFS canonical spec: https://github.com/kryptobaseddev/lafs-protocol/blob/main/lafs.md

MCP Servers

caamp mcp install <source>         # Install MCP server to agent configs
caamp mcp remove <name>            # Remove MCP server from configs
caamp mcp list                     # List configured MCP servers
caamp mcp list -a cursor           # List for a specific agent
caamp mcp detect                   # Auto-detect MCP configurations

Instructions

caamp instructions inject          # Inject blocks into instruction files
caamp instructions check           # Check injection status across providers
caamp instructions update          # Update all instruction file injections

Config

caamp config show <provider>       # Show provider config contents
caamp config path <provider>       # Show config file path
caamp doctor                       # Diagnose configuration and health issues

Advanced (LAFS-compliant wrappers)

caamp advanced providers --min-tier medium --details
caamp advanced batch --mcp-file ./mcp-batch.json --skills-file ./skills-batch.json
caamp advanced conflicts --mcp-file ./mcp-batch.json
caamp advanced apply --mcp-file ./mcp-batch.json --policy skip
caamp advanced instructions --content-file ./AGENT-BLOCK.md --scope project
caamp advanced configure -a claude-code --global-mcp-file ./global-mcp.json --project-mcp-file ./project-mcp.json

Global Flags

| Flag | Description | |------|-------------| | -a, --agent <name> | Target specific agent(s), repeatable | | -g, --global | Use global/user scope (default: project) | | -y, --yes | Skip confirmation prompts | | --all | Target all detected agents | | --json | JSON output format | | --dry-run | Preview changes without writing |

Supported Providers

CAAMP supports 44 AI coding agents across 3 priority tiers:

| Priority | Providers | |----------|-----------| | High | Claude Code, Cursor, Windsurf | | Medium | Codex CLI, Gemini CLI, GitHub Copilot, OpenCode, Cline, Kimi, VS Code, Zed, Claude Desktop, Amazon Q Developer, GitHub Copilot CLI | | Low | Roo, Continue, Goose, Antigravity, Kiro, Amp, Trae, Aide, Pear AI, Void AI, Cody, Kilo Code, Qwen Code, OpenHands, CodeBuddy, CodeStory, Aider, Tabnine, Augment, JetBrains AI, Devin, Replit Agent, Mentat, Sourcery, Blackbox AI, Double, Codegen, SWE-Agent, Forge, Gemini Code Assist |

Config Key Mapping

Each provider uses a different key name for MCP server configuration:

| Config Key | Providers | |------------|-----------| | mcpServers | Claude Code, Cursor, Windsurf, Gemini CLI, GitHub Copilot, Cline, Kimi, and 12 others | | mcp_servers | Codex | | extensions | Goose | | mcp | OpenCode | | servers | VS Code | | context_servers | Zed |

Instruction File Mapping

| File | Providers | |------|-----------| | CLAUDE.md | Claude Code, Claude Desktop | | GEMINI.md | Gemini CLI | | AGENTS.md | All other providers (Cursor, Windsurf, Codex, Kimi, etc.) |

Architecture

┌─────────────────────────────────────────────────┐
│                   CLI Layer                      │
│  providers │ skills │ mcp │ instructions │ config│
├─────────────────────────────────────────────────┤
│                 Core Layer                       │
│  registry │ formats │ skills │ mcp │             │
│  marketplace │ sources │ instructions            │
├─────────────────────────────────────────────────┤
│                Data Layer                        │
│  providers/registry.json │ lock files │ configs  │
└─────────────────────────────────────────────────┘

• Provider Registry: Single providers/registry.json with all provider definitions
• Format Handlers: JSON, JSONC (comment-preserving), YAML, TOML
• Skills Model: Canonical copy + symlinks (install once, link to all agents)
• MCP Transforms: Per-agent config shape transforms for Goose, Zed, OpenCode, Codex, Cursor
• Lock File: Tracks all installations at ~/.agents/.caamp-lock.json

Documentation

Read the full documentation on GitBook

| Document | Description | |----------|-------------| | API Reference | Full library API (signatures and examples) | | Advanced CLI | LAFS-compliant advanced command wrappers and input/output schemas | | Advanced Recipes | Production TypeScript patterns for tier filtering, rollback, conflict handling, and dual-scope operations | | Provider Configuration Guide | Config keys, formats, scopes, and provider mapping guidance | | Migration Guide | Upgrade notes for moving to v1.0.0 | | Troubleshooting | Common failure modes and remediation steps | | CLI Help Examples | --help command examples for every command group | | Skills Recommendations | Marketplace search and recommendation engine | | Contributing | Development workflow and PR expectations | | Security Policy | Private vulnerability disclosure process | | LAFS Compliance Profile | CAAMP-specific LAFS adoption scope and compliance mapping | | Agents Directory Standard | .agents/ standard directory structure | | LAFS Specification | Canonical cross-language LLM-agent-first protocol | | Technical Specification | RFC 2119 spec covering all subsystems | | Vision & Architecture | Project vision, design philosophy, and architecture | | Gap Analysis & Roadmap | Current state vs plan, v0.2.0+ roadmap |

Contributing

Provider definitions live in providers/registry.json. To add a new AI coding agent:

1. Add a provider entry to providers/registry.json with all required fields
2. Run npm test to validate the registry
3. Submit a PR

See the Technical Specification for the full provider schema.

License

MIT