# AgentPlaybooks [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/matebenyovszky/agentplaybooks  
**GitHub Stars:** 9  
**npm Downloads (last month):** 645  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/agentplaybooks

## Description
Portable, vendor-neutral home for agent skills, MCP servers, personas, and memory.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "agentplaybooks": {
    "command": "npx",
    "args": ["-y","@agentplaybooks/cli@latest"],
    "env": {
      "NEXT_PUBLIC_SUPABASE_URL": "",
      "NEXT_PUBLIC_SUPABASE_ANON_KEY": "",
      "SUPABASE_SERVICE_ROLE_KEY": "",
      "SECRETS_ENCRYPTION_KEY": ""
    }
  }
}
```

**Requires environment variables:** `NEXT_PUBLIC_SUPABASE_URL`, `NEXT_PUBLIC_SUPABASE_ANON_KEY`, `SUPABASE_SERVICE_ROLE_KEY`, `SECRETS_ENCRYPTION_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What AgentPlaybooks MCP server does

AgentPlaybooks MCP server provides an MCP-facing way to work with hosted playbooks. A playbook is a portable configuration record for an agent, combining one persona with project instructions, JSON-defined skills, optional skill documentation and attachments, MCP servers, memory, and versioned markdown canvas documents.

The service is designed around a single source of truth that can be used across agent platforms. The repository describes support for Claude, ChatGPT/Codex, Cursor, Gemini, local models, and other platforms through its CLI, plugins, exports, and host-specific files. It also supports public and unlisted playbooks and team access through Viewer, Coworker, and Admin roles.

## How it works

AgentPlaybooks MCP server exposes account-level access to playbooks and their related resources. The documented tool surface includes playbooks, skills, memory, canvas, connected MCP and OpenAPI tools, workflows, and a vault-backed API proxy. The same package can be used as a ChatGPT/Codex plugin or as a Claude Code and Claude Cowork plugin, with a skill, commands where applicable, and account MCP access.

Memory is stored as tagged key-value data with descriptions. Canvas provides versioned markdown work documents scoped to a playbook run. Federation allows external MCP servers and OpenAPI services to become tools on a playbook, while credentials are resolved from the platform's vault by name.

The CLI can audit a local project with `doctor`, create or update host configuration with `sync`, and synchronize a hosted playbook with `login`, `playbooks`, `pull`, and `push`. Plugin export and import support Agent Plugins 1.0 packages. Backups are private and versioned; secret values are not included in plugin packages or transferred during synchronization.

## Setup and configuration

To run the repository locally, install Node.js 22 or newer and npm 11 or newer, clone the project, and run `npm install`. A Supabase project is required. Cloudflare is optional for deployment, while the documented worker deployment uses the project's build and Wrangler commands.

Required environment variables are `NEXT_PUBLIC_SUPABASE_URL`, `NEXT_PUBLIC_SUPABASE_ANON_KEY`, `SUPABASE_SERVICE_ROLE_KEY`, and `SECRETS_ENCRYPTION_KEY`. The encryption key must contain 32 bytes represented as 64 hexadecimal characters. Losing it prevents decryption of stored vault secrets, and rotating it invalidates existing encrypted values.

`NEXT_PUBLIC_APP_URL`, `ALLOWED_ORIGINS`, and `SECRETS_REQUIRE_ALLOWED_HOSTS` are optional settings for canonical URLs, CORS control, and mandatory secret host allow-lists. The local development command is `npm run dev`.

## Tools and capabilities

AgentPlaybooks MCP server supports these documented capability areas:

- Read and manage hosted playbooks.
- Work with skills, including portable skill content and attachments.
- Read and update tagged memory values.
- Create or use run-scoped markdown canvas documents.
- Invoke connected MCP and OpenAPI tools through a playbook.
- Run documented workflows.
- Use a vault-backed API proxy without placing secret values in the agent context.
- Audit and synchronize local agent configuration through the CLI.
- Export or import Agent Plugins, MCP, OpenAPI, JSON, Anthropic, and Markdown formats.

Vault operations and federated calls are recorded in an audit trail. Secrets can have host allow-lists, and agents reference secret names rather than receiving the underlying values.

## Limitations and notes

AgentPlaybooks MCP server is part of a broader application rather than a standalone local utility described with one generic installation command. Hosted use depends on the AgentPlaybooks account and API-key model; self-hosting requires Supabase and the application configuration.

The system does not eliminate host-specific setup: the CLI writes platform files for enabled targets, and compatibility depends on the target's MCP or plugin support. Secret values remain outside portable exports and backups, so moving a playbook transfers references and bindings rather than credentials themselves.

_Full upstream README: https://allmcps.com/mcp/agentplaybooks/readme_