If in the past year you've heard about MCP — Model Context Protocol — and wondered whether yet another protocol was worth learning, the short answer is: yes. Born at Anthropic in late 2024, MCP has become in a matter of months the de facto standard for giving AI assistants access to external tools: databases, files, APIs, code search, organizational knowledge graphs.
The most common analogy is that MCP is "the USB-C of AIs". You write the server once, and you can plug it into Claude, ChatGPT, Cursor, Gemini, n8n — without rewriting anything.
In this article: what MCP actually is, how to configure it, which clients support it today (with the config file path for each one), and how to use it inside an n8n workflow.
What MCP is, in two sentences
An MCP server exposes tools, resources and reusable prompts to a client. The MCP client — usually an AI assistant — calls those tools during the conversation, gets the results, and folds them into its responses.
Three things make it interesting:
- Model-agnostic: the same server works with Claude, GPT-5, Gemini.
- Transport-agnostic: stdio for local processes, SSE/HTTP for remote servers.
- Standardized: the spec is public, official SDKs ship in Python, TypeScript, Go, C#.
Configuring it, in practice
All clients adopt the same idea: a JSON file with the list of MCP servers to enable. They differ only in the file location and a few syntactic details.
The typical configuration of a local MCP server (launched as a subprocess) looks like this:
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]
},
"prowodo": {
"command": "npx",
"args": ["-y", "@prowodo/mcp-client"],
"env": {
"PROWODO_TOKEN": "${PROWODO_TOKEN}"
}
}
}
}
For a remote server (HTTP or SSE) the shape is:
{
"mcpServers": {
"prowodo": {
"url": "https://backend.prowodo.com/mcp/",
"transport": "sse"
}
}
}
Clients that support MCP today
| Client | Type | Where to configure it |
|---|---|---|
| Claude Desktop / Claude Code | AI assistant (Anthropic, the protocol's author) | ~/.config/claude/settings.json (Claude Code), claude_desktop_config.json (Desktop) |
| ChatGPT Desktop & OpenAI Agents SDK | AI assistant (OpenAI) | UI Custom Integrations / SDK config |
| Gemini CLI / Code Assist (VS Code) | AI assistant (Google) | ~/.gemini/settings.json or mcpServers block in project |
| GitHub Copilot in VS Code (Agent Mode) | Native IDE extension | .vscode/mcp.json per project, user settings for global |
| Cursor | AI-native IDE | ~/.cursor/mcp.json or .cursor/mcp.json per project |
| Windsurf (Codeium) | AI-native IDE | ~/.windsurf/mcp.json |
| Cline | VS Code extension | cline_mcp_settings.json |
| Zed editor | IDE | Context server config inside settings.json |
| Continue.dev | IDE extension | ~/.continue/config.yaml |
| JetBrains AI Assistant | JetBrains extension | UI Settings → AI Assistant → MCP |
| Sourcegraph Cody | Code search + AI | Integrations UI |
| Goose (Block) | Agent CLI | ~/.config/goose/profiles.yaml |
In every case the syntax is essentially the one shown above: only the file changes, not the structure.
What about n8n? Yes, that one too
n8n — the most popular open-source automation/workflow tool — got MCP support in 2025, first as a community node and then in core. Today you can:
- Expose an n8n workflow as an MCP server, so Claude, ChatGPT or a Cursor agent can call it as a tool. Example: "read last week's Shopify orders and produce a report" becomes a simple tool call.
- Use an existing MCP server as a tool inside an AI Agent node in n8n. A workflow can query a database via MCP, publish a release on a custom system, update a CRM.
You configure it from the MCP Server Trigger node (to expose) or from the MCP Client Tool node (to consume), with the same JSON structure as the traditional clients.
For people automating business processes this is the natural fit: MCP + n8n turn AI from a "smart chatbot" into an orchestrator with real access to your stack.
Why this matters for us (and probably for you)
ProWoDo exposes its own MCP server at https://backend.prowodo.com/mcp/, OAuth2-authenticated. That means whoever is helping you work — Claude, ChatGPT, a Gemini agent, an n8n workflow — can read and update your team's tasks without touching the UI.
A real example we use ourselves: in the morning you open Claude Code, you say "give me a recap of the open tasks for project X", and the list comes back, pulled straight from ProWoDo. You create a new task with "add to the current sprint: auth bug, high priority", and the task is created. The exact same server works from n8n when you need to automatically track issues coming in from an external form.
It's the same mechanism we use internally to build ProWoDo: full dogfooding. Every feature starts as a task we created via MCP and ends with a commit referencing that same task.
What to do now
- Pick the client you already use and look for the "MCP" entry in its settings.
- Add a first simple server (
filesystemorgitare great starters). - When you need a real use case, create a ProWoDo account and connect our MCP server: managing your team's tasks from any AI assistant changes the way you work.
If you're stuck configuring the client you use, drop us a line. We're collecting tested configurations for the most common cases — we want anyone, not just developers, to be able to use AI as a project-management copilot.