Quickstart

Your first PML workflow in 5 minutes

Prerequisites

Before starting, ensure you have:

• PML installed and configured (Installation Guide)
• At least one MCP server configured (e.g., filesystem)
• Claude Code or another MCP client running

Step 1: Verify PML is Running

If using stdio mode with Claude Code, PML starts automatically when you launch Claude Code.

For HTTP mode, start manually:

./pml serve --config config/mcp-servers.json --port 3001

You should see:

🚀 Starting PML MCP Gateway...
✓ Connected: filesystem
✓ Connected: memory
PML gateway running on port 3001

Step 2: Search for Tools

Let's find tools using semantic search. In Claude Code, ask:

Use pml:search_tools to find tools related to "read files"

PML returns relevant tools ranked by semantic similarity:

Results:
  1. filesystem:read_file (0.95) - Read file contents
  2. filesystem:list_files (0.82) - List directory contents
  3. filesystem:read_directory (0.78) - Read directory tree

The scores indicate how well each tool matches your intent.

Step 3: Execute a Simple DAG

Now let's run a multi-step workflow. Ask Claude:

Use pml:execute_dag with intent "Read package.json and extract the project name"

PML automatically:

1. Identifies the required tools
2. Builds a DAG with proper dependencies
3. Executes the workflow

Result:

{
  "status": "complete",
  "results": {
    "read": { "content": "{ \"name\": \"my-project\", ... }" },
    "extract": { "name": "my-project" }
  },
  "metrics": {
    "total_time_ms": 145
  }
}

Step 4: Run Code in the Sandbox

For data processing, use the sandbox. Ask:

Use pml:execute_code to count lines in a JSON array:
[{"name": "Alice"}, {"name": "Bob"}, {"name": "Charlie"}]

PML executes the code in an isolated Deno environment:

// PML executes this safely
const data = [{ "name": "Alice" }, { "name": "Bob" }, { "name": "Charlie" }];
return data.length; // Returns: 3

Step 5: Parallel Execution

Watch PML parallelize independent tasks. Ask:

Use pml:execute_dag with intent "Read config.json, package.json, and README.md simultaneously"

PML detects these are independent and runs them in parallel:

Layer 0: [read_config, read_package, read_readme]  ← PARALLEL
         Time: 89ms (not 3 × 89ms = 267ms)

What You've Learned

Next Steps

Now explore more advanced features:

• Concepts - Deep dive into how PML works
• DAG Execution - Complex workflows
• Capabilities - Learned patterns
• MCP Tools Reference - Complete API documentation

Common Issues

PML not responding

Check if the gateway is running:

./pml status

Tools not found

Re-initialize to refresh tool discovery:

./pml init --config config/mcp-servers.json

Slow execution

Enable the cache for repeated operations:

./pml serve --config ... # Cache enabled by default