Modern programming tools like Cursor, Windsurf, Augment Code, and Cline all feature powerful "agent modes" that can automatically complete complex tasks based on natural language instructions:
- Updating package dependencies and addressing CVE vulnerabilities
- Performing routine code refactoring across multiple files
- Adding unit tests for new or existing functionality
- Standardizing code formatting and fixing linting issues
However, these capabilities face a critical limitation: context sharing.
- Each team member uses different tools (Cursor, Cline, etc.)
- Natural language "playbooks" exist only in individual environments
- No centralized way to share, version, or collaboratively improve these playbooks
- Playbook silos form as team members develop their own agent instructions
playbookmcp converts GitHub repository-based playbooks into an MCP Server that all mainstream programming tools support. This enables:
- Collaborative editing of playbooks through standard GitHub workflows
- Version control for your team's playbooks
- Immediate sharing of new capabilities across the entire team
For example, when engineer Alice adds a new playbook for "Updating dependency versions across the monorepo," commits it to the playbook repo, and pushes, engineer Bob immediately gains access to this capability in his preferred coding tool.
Your team can continuously enrich the knowledge base by editing the playbook-repo (typically a GitHub repository), enabling AI to automatically complete various tasks.
For playbook repository structure details, see docs/playbook.md.
To use playbookmcp with a remote playbook repository:
{
"mcpServers": {
"playbookmcp": {
"command": "npx",
"args": [
"-y",
"playbookmcp@latest",
"--playbook-repo",
"<your playbook repo>"
]
}
}
}For testing with a local context directory:
{
"mcpServers": {
"playbookmcp-local": {
"command": "npx",
"args": [
"-y",
"playbookmcp@latest",
"--playbook-path",
"<your playbook files path>"
]
}
}
}For developing PlaybookMCP itself:
{
"mcpServers": {
"playbookmcp-dev": {
"command": "node",
"args": [
"build/index.js",
"--playbook-path",
"<your playbook files path>"
]
}
}
}You can also use inspector to inspect the MCP resources:
npx @modelcontextprotocol/inspector node build/index.js --playbook-repo <your playbook repo>Below is an example configuration in Cursor:
{
"mcpServers": {
"playbookmcp": {
"command": "npx",
"args": [
"-y",
"playbookmcp@latest",
"--playbook-repo",
"[email protected]:xuezhaojun/PlaybookMCP-demo.git"
]
}
}
}The first command must be pb_start to initiate interaction:
pb_start Say Hi to playbookmcp.
Or more specific:
use MCP tool pb_start Say hi to playbookmcp.
Your code-agent will then add a comment on this issue.

