Ductape MCP Server

The Ductape MCP Server is a Model Context Protocol (MCP) server that exposes the full Ductape SDK as tools for AI agents and copilots. Connect it to Cursor, Claude Desktop, or any MCP-compatible client and let an LLM query databases, run actions, manage storage, orchestrate workflows, and more — without embedding the SDK in your agent process.

What it does

Capability	Description
SDK execution	Call any allowed SDK module and method through the Ductape backend proxy
Payload generation	Produce canonical executable payload templates with schema metadata
Code snippets	Generate ready-to-copy TypeScript or Python SDK examples

All operations go through https://api.ductape.app. The MCP process is stateless — it never loads the SDK locally and holds no workspace credentials between requests.

Architecture

Key properties:

• Stdio transport — The server communicates over stdin/stdout. An MCP client spawns it as a child process.
• Per-request authentication — Every ductape_execute call requires a publishable_key. No keys are stored in the server process.
• Proxy-enforced allowlist — Only module/method pairs validated by the backend proxy can run. Unknown calls are rejected before execution.
• No environment variables required — The server starts with zero configuration; credentials are passed per tool invocation.

When to use it

Use the MCP server when you want an AI agent to:

• Operate your Ductape workspace — Create products, connect apps, configure environments, manage components.
• Run runtime operations — Query databases, upsert vectors, upload files, send notifications, dispatch actions.
• Build integration code — Generate typed payloads and SDK snippets for engineers from natural-language intent.
• Automate platform tasks — Schedule jobs, manage sessions, check health probes, run quota/fallback chains.
• Debug and inspect — List resources, fetch dashboards, query logs, check job status.

You do not need the MCP server if you are writing application code directly with @ductape/sdk. Use the SDK in your backend; use MCP when an external AI client needs programmatic access to Ductape.

Tools at a glance

Tool	Purpose
ductape_execute	Run any allowed SDK operation
ductape_generate_payload	Generate canonical payload + schema metadata
ductape_generate_snippet	Generate payload + TypeScript/Python snippet

Supported SDK modules

The server exposes 20 SDK modules covering the full Ductape platform:

product, app, databases, graph, webhooks, notifications, messageBrokers, storage, vector, caches, sessions, quotas, actions, features, jobs, logs, resilience, health, fallback, secrets

See Modules & methods for the complete allowlist and Method reference for parameter signatures.

Documentation map

• Getting started — Install, build, and verify the server
• Configuration — Cursor, Claude Desktop, and other MCP clients
• Tools reference — Arguments, responses, and examples for all three tools
• Modules & methods — Allowed modules and method names
• Method reference — Exhaustive parameter signatures per method
• Code generation — Payload and snippet generation workflows
• Use cases & possibilities — What agents can accomplish across the platform
• Security — Authentication model and multi-tenant safety

Source

The MCP server lives in the monorepo at mcp-server/. Package name: @ductape/mcp-server.