MCP — OneHub Design

What is MCP

The Model Context Protocol is a lightweight, open protocol that standardizes how AI applications connect to external systems. Instead of building custom integrations for every tool, MCP provides a universal interface that any AI model can use to interact with real-world capabilities.

Read Data

Access external sources — files, databases, APIs, documents — through a single protocol. One MCP server can give your AI access to an entire ecosystem of data.

Execute Actions

Create records, send messages, run commands, and trigger workflows in connected systems. MCP standardizes how AI models invoke tools across any environment.

Discover Capabilities

Automatically detect what tools are available and how to use them. MCP servers declare their own capabilities — no manual configuration needed.

How MCP Works in OneHub

OneHub acts as both an MCP client and an MCP host. This gives you the flexibility to connect to external tools and expose your own capabilities.

As a Client

OneHub connects to external MCP servers to give its AI models access to additional tools and data. Connected server tools become available in:

AI chat — Use MCP tools directly in conversation. Query databases, check deployment status, and more.
Automations — Use MCP tools as steps in your automation workflows. Connect them to triggers, conditions, and other actions.
Design workflows — Give your design AI access to asset libraries, brand databases, and content repositories through MCP.

As a Host

OneHub exposes its own capabilities as MCP tools, so other MCP-compatible applications can use OneHub as a tool provider. Connect OneHub to Claude Desktop, Cursor, Windsurf, or any other MCP client:

Query project data — Read files, search content, and access project metadata.
Manage resources — Create, update, and delete files in your OneHub workspace.
Run automations — Trigger OneHub automations from external applications.
Access media — Generate images, videos, and audio through OneHub's media pipeline.

Connecting MCP Servers

Add a Server

1 Open Settings → MCP in OneHub and click Add Server.
2 Enter the server URL or command. OneHub supports HTTP servers, local processes, and SSE servers for real-time communication.
3 Authenticate if required. OneHub supports API keys, OAuth, and custom headers.
4 Click Connect. OneHub discovers available tools automatically.
5 Start using those tools in AI chat, automations, and design workflows.

Manage Your Servers

Tool Discovery

OneHub automatically detects all tools exposed by each MCP server and makes them available in the AI and automation interfaces.

Per-Server Permissions

Control which tools are enabled and what data each server can access. Fine-grained permission management per connection.

Health Monitoring

OneHub tracks server availability and alerts you when a connection drops. Automations pause and resume automatically.

Usage Logs

See which tools are being called, how often, and what data is flowing through them. Full observability into MCP activity.

Built-in MCP Server

OneHub ships with a built-in MCP server that exposes OneHub's own features as tools. Connect this server to any MCP-compatible client to unlock OneHub's capabilities from external applications.

Query Project Data

Read files, search content, and access project metadata from any MCP-compatible client.

Manage Resources

Create, update, and delete files in your OneHub workspace through MCP tool calls.

Run Automations

Trigger OneHub automations from external applications. Build cross-platform workflows.

Access Media

Generate images, videos, and audio through OneHub's media pipeline via MCP.

Query Brand Memory

Access your brand settings, design tokens, and style preferences from any connected tool.

MCP Use Cases

AI-Powered Development

Connect OneHub to Claude Desktop or Cursor via MCP. Your AI coding assistant gets access to project files, brand assets, and design context — generating code that matches your brand and fits your architecture.

Unified Tool Access

Connect your tools to MCP servers once. Every MCP-compatible application — OneHub, Claude Desktop, Cursor, Windsurf, and others — gets access to the same tools with the same permissions.

Data-Driven Automation

Use MCP tools as building blocks in OneHub automations. Query a database, transform results with AI, write to a spreadsheet, and notify your team — all through MCP-connected tools.

Brand-Aware Generation

Connect your design system, asset library, or brand database as an MCP server. OneHub's AI can pull real brand assets and ensure consistency without specifying every detail.

Cross-Application Workflows

Chain tools across applications through MCP. Start a workflow in OneHub, hand off to a remote MCP server, get results back, and continue — all through the same protocol.

MCP Server Examples

Popular MCP servers you can connect to OneHub:

Filesystem Read and write files

GitHub Repos, issues, PRs

PostgreSQL Database queries

Slack Messages and channels

Notion Pages and databases

Browser Web control and testing

Memory Persistent key-value storage

Brave Search Web search

Google Maps Location and directions

Stripe Payments and subscriptions

Cloudflare Workers, Pages, DNS

Building Your Own MCP Server

OneHub's MCP support is fully compatible with the MCP specification. If you want to expose your own tools, follow these steps:

Implement the Protocol

Use one of the official SDKs — TypeScript, Python, or Go — to build an MCP server.

Define Your Tools

Declare each tool with a name, description, and input schema using JSON Schema for parameter definitions.

Host Your Server

Run it locally, deploy to a cloud provider, or package it as a Docker container.

Connect to OneHub

Add the server URL or command in OneHub's MCP settings. Your tools appear automatically.

Security and Permissions

Tool-Level Permissions

Enable or disable individual tools from each MCP server. Granular control over what each connection can do.

Data Isolation

Each MCP connection operates in its own context. Tools from one server cannot access data from another unless explicitly allowed.

Credential Storage

API keys and authentication tokens are encrypted at rest and never exposed in logs or automation outputs.

Audit Logging

Every MCP tool call is logged with the requesting application, tool name, parameters, and result status.

FAQ

What is MCP vs a plugin?

MCP is a protocol, not a plugin. It standardizes how AI applications communicate with external tools. Any MCP server works with any MCP client — there is no vendor lock-in.

Does MCP work with all AI models?

MCP works with any AI model that supports tool use. OneHub routes MCP tool calls through whichever model you are using — GPT-4, Claude, Gemini, or any other supported provider.

Can I run MCP servers locally?

Yes. OneHub supports local MCP servers that run as processes on your machine. Your data never leaves your device.

What if an MCP server goes down?

OneHub monitors server health and alerts you when a connection drops. Automations that depend on unavailable servers pause automatically and resume when the connection is restored.

Model Context Protocol