MCP Server¶

pjkm exposes its full scaffolding engine as an MCP (Model Context Protocol) server. AI agents can create, preview, and manage Python projects programmatically.

Install¶

pip install pjkm[mcp]

Run¶

python -m pjkm.mcp                                    # stdio (Claude Desktop)
pjkm-mcp                                              # same thing
fastmcp run pjkm.mcp.server:mcp --transport sse       # SSE (remote agents)

Architecture¶

        graph TB
    subgraph "Clients"
        CD[Claude Desktop]
        LC[LangChain Agent]
        CU[Custom Agent]
    end
    subgraph "pjkm MCP Server"
        T1[init_project]
        T2[add_groups]
        T3[preview_project]
        T4[list_recipes]
        T5[list_groups]
        T6[get_group_info]
        T7[search_registry]
        T8[adopt_project]
        R1["pjkm://recipes"]
        R2["pjkm://groups"]
        R3["pjkm://groups/{id}"]
        R4["pjkm://registry"]
    end
    subgraph "pjkm Core"
        ENG[ProjectEngine]
        REG[GroupRegistry]
        RES[GroupResolver]
        TPL[TemplateComposer]
    end
    CD --> T1 & T4 & T5
    LC --> T1 & T2 & T7
    CU --> T3 & T6 & T8
    T1 --> ENG
    T2 --> REG
    T5 --> REG
    T6 --> REG
    ENG --> RES --> TPL

Tools¶

`init_project`¶

Create a new project from a recipe or custom groups.

init_project(name="my-api", recipe="fastapi-service")
init_project(name="my-lib", archetype="single-package", groups=["dev", "docs"])

`add_groups`¶

Add groups to an existing project.

add_groups(groups=["auth", "redis"], directory="./my-api")

`preview_project`¶

See what would be generated without creating files.

preview_project(recipe="saas-backend")

`list_recipes`¶

Browse all 22 recipes with descriptions.

`list_groups`¶

Browse all 105 groups, optionally filtered by category.

list_groups(category="AI / ML")

`get_group_info`¶

Detailed info for a group: deps, scaffolded files, config.

get_group_info(group_id="database")

`search_registry`¶

Search community packs.

search_registry(query="django")

`adopt_project`¶

Scan an existing project and detect frameworks/tools.

adopt_project(directory="./my-old-project")

Resources¶

URI	Description
`pjkm://recipes`	All recipes with full details
`pjkm://groups`	All groups organized by category
`pjkm://groups/{id}`	Detailed info for a specific group
`pjkm://registry`	Community pack listing

Claude Desktop Setup¶

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "pjkm": {
      "command": "python",
      "args": ["-m", "pjkm.mcp"]
    }
  }
}

Then ask Claude: “Create a FastAPI service with database and auth”

LangChain Agent Setup¶

from langchain_mcp_adapters import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent
from langchain_anthropic import ChatAnthropic

model = ChatAnthropic(model="claude-sonnet-4-20250514")

async with MultiServerMCPClient({
    "pjkm": {"command": "python", "args": ["-m", "pjkm.mcp"]}
}) as client:
    tools = client.get_tools()
    agent = create_react_agent(model, tools)
    result = await agent.ainvoke({
        "messages": [{"role": "user", "content": "Create an AI agent project with LangGraph"}]
    })

Use Cases¶

Agent Sandbox Bootstrapping¶

An orchestrator agent creates isolated project environments for task-specific agents:

        sequenceDiagram
    participant O as Orchestrator
    participant P as pjkm MCP
    participant A as Task Agent
    O->>P: init_project("task-123", recipe="ai-agent")
    P-->>O: Created at /sandboxes/task-123/
    O->>A: Execute in /sandboxes/task-123/
    A->>A: Uses generated agent scaffolding
    A-->>O: Task complete

Multi-Agent Platform Setup¶

Create a full platform with shared infrastructure:

# Agent creates a workspace
init_project("my-platform", ...)  # Not quite — use workspace CLI
# But can create individual services:
init_project("api-gateway", recipe="fastapi-service", directory="platform/")
init_project("ml-worker", recipe="ai-agent", directory="platform/")
init_project("data-ingestion", recipe="etl-pipeline", directory="platform/")