Code Mode

Write Python or TypeScript scripts for bulk workflows and programmatic tool calling

Code mode lets you write local scripts that call MCP tools through the DataGen SDK. Use it when you need batch processing, complex logic, or production-grade automation that goes beyond interactive MCP tool calls.

Using Claude Code? Run /datagen:code-mode for guided scripting with tool discovery, durable checkpoint patterns, and large output handling.

When to Use Code Mode

Use MCP Tools When

• Interactive discovery and debugging
• Simple one-off operations
• Learning tool schemas
• AI agent-assisted work

Use Code Mode When

• Batch processing (100+ operations)
• Scheduled jobs and cron tasks
• CI/CD pipelines
• Complex data transformations
• Production applications

Rule of thumb: fewer than 5 tool calls with small results — use MCP directly. More than 5 calls, large data, or complex logic — write a script with the SDK.

---

Installation

Python

pip install datagen-python-sdk

TypeScript

npm install @datagen-dev/typescript-sdk

Set your API key:

export DATAGEN_API_KEY="your_api_key_here"

---

Quick Start

Python

from datagen_sdk import DatagenClient

client = DatagenClient()

# Call any MCP tool as a Python function
result = client.execute_tool(
    "mcp_Linear_list_issues",
    {"filter": {"state": {"name": {"eq": "In Progress"}}}}
)

print(f"Found {len(result)} active issues")

TypeScript

import { DatagenClient } from '@datagen-dev/typescript-sdk';

const client = new DatagenClient();

const result = await client.executeTool(
    "mcp_Linear_list_issues",
    { filter: { state: { name: { eq: "In Progress" } } } }
);

console.log(`Found ${result.length} active issues`);

---

Patterns

Simple Script

One-off data operations:

from datagen_sdk import DatagenClient

client = DatagenClient()

# Fetch leads from database
leads = client.execute_tool("mcp_Supabase_run_sql", {
    "params": {
        "sql": "SELECT * FROM leads WHERE score > 80",
        "projectId": "your-project",
        "databaseName": "your-db"
    }
})

# Create follow-up tasks in Linear
for lead in leads:
    client.execute_tool("mcp_Linear_create_issue", {
        "title": f"Follow up with {lead['company']}",
        "description": f"Contact: {lead['email']}\nScore: {lead['score']}",
        "teamId": "your-team-id"
    })

# Notify team on Slack
client.execute_tool("mcp_Slack_chat_postMessage", {
    "channel": "#sales",
    "text": f"Created {len(leads)} follow-up tasks for high-priority leads"
})

Batch Processing with Error Handling

Process large datasets with retry logic:

from datagen_sdk import DatagenClient, DatagenToolError

client = DatagenClient(retries=3, backoff_seconds=1.0)

contacts = [...]  # Your contact list

results = []
errors = []

for contact in contacts:
    try:
        result = client.execute_tool("mcp_Gmail_gmail_send_email", {
            "to": contact["email"],
            "subject": "Welcome!",
            "body": f"Hi {contact['name']}, thanks for signing up!"
        })
        results.append({"email": contact["email"], "status": "sent"})
    except DatagenToolError as e:
        errors.append({"email": contact["email"], "error": str(e)})

print(f"Sent: {len(results)}, Failed: {len(errors)}")

Deploy as Custom Tool

Turn your script into a deployed API endpoint:

from datagen_sdk import DatagenClient

client = DatagenClient()

# Deploy your script as a custom tool
deployment = client.deploy_custom_tool(
    name="weekly_lead_enrichment",
    description="Enrich new leads with LinkedIn and web data",
    code=open("enrich.py").read(),
    input_schema={
        "type": "object",
        "properties": {
            "domains": {"type": "array", "items": {"type": "string"}}
        },
        "required": ["domains"]
    },
    output_vars=["enriched_leads", "total"],
    expected_tools=["mcp_LinkedIn_get_company", "mcp_Perplexity_search"],
    mcp_server_names=["LinkedIn", "Perplexity"]
)

print(f"Deployed: {deployment['deployment_uuid']}")

# Run it
result = client.run_custom_tool(
    deployment["deployment_uuid"],
    input_vars={"domains": ["acme.com", "globex.com"]}
)
print(result["output_vars"])

---

Error Handling

from datagen_sdk import (
    DatagenClient,
    DatagenAuthError,
    DatagenToolError,
    DatagenHttpError
)

client = DatagenClient(retries=3, backoff_seconds=0.5)

try:
    result = client.execute_tool("mcp_Linear_list_projects", {"limit": 10})
except DatagenAuthError:
    print("Authentication failed - check your API key")
except DatagenToolError as e:
    print(f"Tool execution failed: {e}")
except DatagenHttpError as e:
    print(f"HTTP error: {e}")

Error	Description
DatagenAuthError	Authentication failed (401/403) — check API key or MCP connection
DatagenToolError	Tool execution failed — check parameters
DatagenHttpError	Network or HTTP-level errors
DatagenDeploymentError	Deployment or custom tool run failed
DatagenSecretError	Secret management errors

---

Full API Reference

For complete method signatures, parameters, and return types, see:

Python SDK Reference

Every method, every parameter

TypeScript SDK Reference

Every method, every parameter