# Patchworks MCP

## Introduction

*MCP (Model Context Protocol)* is an open protocol that enables secure, standardised connections between Al assistants (e.g. Claude, Gemini, Chatgpt) and external data sources or tools (such as Patchworks). 

The Patchworks *MCP server* acts as a bridge between an Al assistant and Patchworks. It allows the Al to interact with Patchworks, with precise control over what it can see and do. At a high level:

* MCP uses a client-server model, where an AI agent (the client) makes requests to an MCP server. 
* The MCP server defines and exposes `tools` (e.g. get flow runs, summarise failed run, triage latest run failures) and `resources` (e.g. documentation, databases). 
* AI agents use these `tools` and `resources` to perform actions, thereby expanding their abilities beyond their core language understanding. 

 You can see this in the illustration below:

<figure><img src="https://2440044887-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FLYNcUBVQwSkOMG6KjZfz%2Fuploads%2FLrsLBPLRSfUj1N1NKcq8%2Fmcp%20how%20it%20works.png?alt=media&#x26;token=7ce94803-6b63-4c52-a877-4b02df3b30b9" alt=""><figcaption></figcaption></figure>

This opens the door to a range of possibilities. Your AI assistant can interact with Patchworks directly to triage issues, generate reports, and even run flows - all in natural language. Whether you’re a merchant, partner, or developer, Patchworks MCP transforms how you work with your integrations.

## Why Patchworks MCP?

* **AI-ready iPaaS**\
  Make integrations conversational and intelligent.
* **Faster troubleshooting**\
  Automate triage & identify solutions
* **Customisable**\
  Add your own tools alongside our pre-loaded set
* **Safe & secure** \
  Per-tenant isolation, role-based access, and auditable tool calls.
* **Future-proof**\
  Works with Claude, Gemini, ChatGPT, and any MCP-compatible client.

## Implementation

We provide two implementation paths for the Patchworks MCP - *local* and *hosted*:

<table><thead><tr><th width="121.12109375" valign="top">Type</th><th valign="top">Implementation</th><th valign="top">Pros</th><th valign="top">Cons</th></tr></thead><tbody><tr><td valign="top"><a href="patchworks-mcp/patchworks-mcp-local">Local</a></td><td valign="top">Clone a Patchworks MCP repository to retrieve setup files, then install dependencies before configuring your environment and MCP installation.</td><td valign="top">Complete control over the hosting environment.</td><td valign="top">Technical implementation and ongoing maintenance.<br><br>Default tools cannot be customised.</td></tr><tr><td valign="top"><a href="patchworks-mcp/patchworks-mcp-hosted">Hosted</a></td><td valign="top">Minimal setup outside of the Patchworks dashboard and your AI assistant.</td><td valign="top">Less technical implementation; automatic updates;  customise and deploy tools via the Patchworks dashboard. </td><td valign="top">No control of the hosting environment.<br> </td></tr></tbody></table>

{% hint style="info" %}
Our product documentation can also be integrated with AI assistants via an MCP server! For details, please refer to the [Patchworks product documentation MCP server](https://doc.wearepatchworks.com/product-documentation/developer-hub/patchworks-mcp/patchworks-product-documentation-mcp-server) section.
{% endhint %}

## Demo

Before looking at MCP in more detail, let's see how it can work in practice. The video below shows Claude identifying failed process flow runs from a given time period, summarising those flow failures, identifying failure patterns, recommending possible solutions, retrying failed runs, and reporting back on the outcome.&#x20;

{% embed url="<https://www.youtube.com/watch?v=XECRW0TEkUE>" %}