Understanding MCP Servers
In the dashboard this appears under MCP servers; internally we still use the bundle data model.
MCP servers are the core organizing principle of MCPBundles. This page explains what they are, how they're categorized, and their lifecycle from creation to use.
What is an MCP server?
An MCP server is a cloud-hosted collection of related tools exposed through a single MCP URL.
Think of it like an app store, but instead of downloading apps, your AI discovers tools. When you add an MCP URL to ChatGPT or Cursor, all the tools in that MCP server become available to your AI.
You may have heard of .mcpb files—the open packaging format for distributing MCP servers (a ZIP archive plus manifest.json). Some clients, including Claude Desktop, can install a server by opening an .mcpb file. On MCPBundles, you connect to the same hosted server with its MCP URL. See the FAQ and the .mcpb format reference for details.
Why MCP servers?
Without them, you'd need to:
- Add 50+ individual tool URLs to your AI
- Remember which tool does what
- Manage credentials for each tool separately
With an MCP server, you:
- Add one URL and get many tools
- Tools are organized by purpose (marketing, development, etc.)
- The connected accounts on the MCP server cover all of its tools
MCP server properties
Every MCP server has:
Basic Information
- Name: What the MCP server is called (e.g., "Marketing Automation")
- Description: What it does and what tools it contains
- Slug: URL-friendly identifier (e.g.,
marketing-automation-a3f) — used in paths likehttps://mcp.mcpbundles.com/bundle/{bundle-slug} - Version: Semantic version number (e.g.,
1.0.0) - Author: Who created the MCP server
- Icon: Visual identifier
Organization
- Category: What type of work the MCP server helps with (see below)
- Tools: List of tools included
- Providers: Which external APIs the tools connect to
Lifecycle Status
- Publication Status: Draft, Published, Deprecated, or Archived
- Readiness Status: Whether you can use it right now (Ready, Needs setup, Needs attention, Disabled, Status unknown)
Access Control
- User Access: Whether you've enabled this MCP server on your account
- Publication: Whether the MCP server is public or private
Categories
MCP servers are organized into categories based on the type of work they help with. This makes it easier to find MCP servers that solve your specific needs.
Available Categories
| Category | Description | Example Tools |
|---|---|---|
| Analytics | Data analysis and reporting | Fetch metrics, generate reports, visualize trends |
| Backend | Server and infrastructure | Deploy code, manage databases, monitor services |
| Cloud | Cloud platform services | AWS, Azure, GCP operations |
| Collaboration | Teamwork and communication | Slack, Teams, Discord integrations |
| Customer Relation | CRM and customer management | HubSpot, Salesforce, Zendesk tools |
| Developer Tools | Coding and development | GitHub, GitLab, code analysis |
| Ecommerce | Online shopping and sales | Shopify, Stripe, inventory management |
| Education | Learning and training | Course management, quiz tools |
| Email marketing and management | Campaign tools, inbox automation | |
| Enterprise | Business operations | ERP, workflow automation |
| Entertainment | Media and content | Streaming, gaming, content creation |
| Financial | Money and accounting | Banking APIs, payment processing |
| Forms | Data collection | Survey tools, form builders |
| Hosting | Web hosting and servers | Domain management, hosting control |
| IoT | Internet of Things | Device control, sensor data |
| Location | Maps and geolocation | Address lookup, routing, geocoding |
| Machine Learning | AI and ML operations | Model training, inference |
| Marketing | Marketing automation | Email campaigns, social media, analytics |
| Media | Image, video, audio | Processing, conversion, editing |
| Messaging | Chat and messaging platforms | SMS, WhatsApp, Telegram |
| Monitoring | System monitoring | Uptime checks, alert management |
| Open Data | Public datasets | Government data, research data |
| Payment | Payment processing | Credit cards, digital wallets |
| Project Management | Task and project tracking | Asana, Trello, Jira tools |
| Search | Search and discovery | Elasticsearch, search APIs |
| Security | Security and authentication | Auth tools, security scanning |
| Social | Social media platforms | Twitter, LinkedIn, Facebook tools |
| Storage | File storage and management | S3, Dropbox, Google Drive |
| Support | Customer support | Ticketing, help desk |
| Telecom | Phone and SMS services | Twilio, voice APIs |
| Text | Text processing | Translation, summarization |
| Time Management | Scheduling and calendar | Calendar APIs, scheduling tools |
| Tools | General utilities | Converters, validators, generators |
| Transport | Shipping and logistics | Delivery tracking, route planning |
| Other | Miscellaneous tools | Tools that don't fit other categories |
Choosing a Category
When creating an MCP server, pick the category that best describes the primary use case. If an MCP server has tools from multiple categories, choose the one that represents the main purpose.
Example: A "Sales Automation" MCP server might include email tools (Email category) and CRM tools (Customer Relation category). Choose Customer Relation since that's the primary purpose.
Publication Status
MCP servers move through different publication states as you work with them.
Draft
- You're still building and testing
- Not visible to others
- You can add/remove tools freely
- Tools won't appear in AI clients
When to use: Creating a new MCP server, major changes to existing ones
Published
- MCP server is complete and ready to use
- Visible in the public directory at
/skills/…(if you've published it) - Can be enabled on your account
- Tools appear in your AI clients
- You can still edit it, but changes are visible immediately
When to use: Everything is tested and ready for use
Deprecated
- Outdated but still works
- Indicates users should migrate to a newer version
- Still functional for existing users
- Not promoted to new users
When to use: You've created a better version and want to sunset the old one
Archived
- No longer maintained
- Hidden from the public directory
- Existing users can keep using it
- New users can't enable it
When to use: Obsolete or no longer supported
Readiness
Readiness tells you whether tools will work for you in this workspace. It reflects workspace access plus which provider connections are bound and verified.
| Label | Typical message | What to do |
|---|---|---|
| Ready | Connections verified | Use tools in your AI or Studio |
| Needs setup | Connect N connections… | Use Connect on the server or add connections in the Credentials section |
| Needs attention | Reconnect or fix scopes | Reconnect or edit the affected connection |
| Disabled | Server turned off | Connect again when you want tools back |
| Status unknown | Could not evaluate | Refresh; open the server detail page |
The list on MCP Servers uses filters such as All, In workspace, and Directory (not the older Connected / Discover split).
Creating and managing MCP servers
Creating an MCP server
- Go to Dashboard → MCP servers
- Click "Create" (or equivalent for a new MCP server)
- Fill in:
- Name (clear, descriptive)
- Description (what it does)
- Category (from the list above)
- Version (start with
1.0.0)
- Click "Create"
- Add tools
- Test with your own credentials
- Publish when ready
Testing tools
Before using tools in your AI, you can test them in Studio at https://www.mcpbundles.com/w/{workspace}/studio — an interactive execution environment that lets you run tools directly in your browser with proper credential context.
Studio features:
- Execute tools with auto-generated forms
- See real-time results and errors
- Get help from an AI assistant
- Verify credentials are working
- Learn tool behavior safely
See the Studio guide for a complete walkthrough.
Adding tools
Tools can belong to multiple MCP servers. This lets you:
- Create general-purpose MCP servers (all marketing tools)
- Create focused MCP servers (only email campaign tools)
- Mix and match tools for specific workflows
To add tools:
- Open the MCP server
- Click "Add Tool"
- Search for existing tools or create new ones
- Tool appears immediately
Enabling on your account
Before you can use the tools, you must enable the MCP server:
- Open MCP Servers → Directory (or search) and pick the server
- Click Connect (or the primary action shown on the row)
- System checks which providers you need
- Scroll to the "Credentials" section
- Click Connect on each provider row that still needs setup in Connections
- Add credentials through the credential panel
- Validate your credentials
- Status changes to Ready when connections are verified
- Scroll to "Quick Setup" section and copy the MCP URL
- Add to your AI client
Disabling
If you stop using an MCP server:
- Open its detail page
- Click Turn off in the header (or remove access if it's a shared server you don't own)
- Confirm in the dialog
- Tools disappear from your AI
- Credentials remain (in case you re-enable later)
- It stays in your list (disabled state)
Accounts and dedicated servers
An MCP server can use more than one account for the same service. For example, you can attach two Stripe accounts to one Stripe server. Every account on that server shares the same MCP URL and the same tool list.
There are two ways to run the same service with different accounts:
| You want | Use | What it looks like |
|---|---|---|
| One URL, and your AI picks the account | Attach more than one account to the server | One URL under Accounts on this server |
| One URL per client or project, each with a fixed account | Dedicated server | A separate MCP server with its own name, URL, and connection, listed under Dedicated servers on the parent server |
A dedicated server doesn't see the accounts attached to the parent server. Each one has only the accounts you attach to it.
Composed MCP servers
You can compose MCP servers by including one inside another. This lets you build custom endpoints that combine tools from multiple sources.
How composition works
When you include MCP server A inside MCP server B:
- All of A's tools become available through B's MCP URL (one
bundle_idfor the client) - Connections are on B — tool calls use credential bindings on the composed server, not bindings you only set on A's catalog page
- After an include, the platform may auto-bind one verified credential per provider onto B; if you have several verified credentials, pick bindings on B manually
- The composed server is a single MCP endpoint with combined tools; templates stay linked by reference, not copied
When to compose
Use composition when:
- You want tools from multiple providers in one endpoint
- You're building a project-specific server (e.g., "Sales Stack" = HubSpot + Resend + Stripe)
- You want a single MCP URL instead of multiple URLs
Keep MCP servers separate when:
- You want to enable/disable groups independently
- Different AI clients need different tool sets
- You're using the Hub endpoint (which already aggregates everything)
Creating a composed MCP server
Composition links template servers into a private server you can edit. Configure Connections on the private server after includes — that is what the composed MCP URL uses at runtime.
Self-serve add-include controls on the server page are partially parked (see Composing MCP servers for what is visible today). You can still view and remove includes on existing private servers.
A dedicated server is composition with exactly one child template (separate URL and connections). Multi-account rules are the same as for any server — see Accounts and dedicated servers above.
See Creating custom MCP servers and Composing MCP servers.
MCP URLs
Each MCP server has a unique MCP URL you add to your AI clients.
Format:
https://mcp.mcpbundles.com/bundle/{bundle-slug}
Example:
https://mcp.mcpbundles.com/bundle/marketing-automation-a3f
Using MCP URLs
Different AI clients configure MCP servers differently:
ChatGPT:
- Settings → Apps → Create app → Paste URL (requires Developer mode)
Cursor:
.cursor/mcp.json→ Add tomcpServers→ Paste URL
Claude Desktop:
- Add Custom Connector → Paste URL
Claude Code:
claude mcp add --transport http {name} {url}
See Integration Guides for detailed steps.
Versioning
MCP servers use semantic versioning: MAJOR.MINOR.PATCH
- MAJOR (1.x.x): Breaking changes, incompatible with previous version
- MINOR (x.1.x): New features, backward compatible
- PATCH (x.x.1): Bug fixes, backward compatible
When to Increment
- Add tools: Increment MINOR (1.0.0 → 1.1.0)
- Remove tools: Increment MAJOR (1.1.0 → 2.0.0)
- Fix tool bugs: Increment PATCH (1.1.0 → 1.1.1)
- Change tool behavior: Increment MAJOR or MINOR depending on impact
Managing Multiple Versions
You can have multiple versions:
marketing-suite-v1(version 1.5.0, deprecated)marketing-suite-v2(version 2.0.0, published)
Users on v1 keep working. New users get v2. Migration is optional.
Common Workflows
Workflow 1: Using an existing MCP server
- Open MCP Servers → Directory or In workspace
- Click Connect on the server detail page (or Turn off to disconnect)
- Connect required providers in the Connections section
- Validate credentials
- Copy the MCP URL from Quick Setup
- Add to your AI client
- Start using tools
Workflow 2: Creating a custom MCP server
- Click "Create"
- Name it based on your workflow (e.g., "Daily Standup Automation")
- Pick a category
- Add relevant tools:
- Slack notification
- Calendar fetch
- Email send
- GitHub PR list
- Test with your credentials
- Publish
- Add to your AI
Workflow 3: Combining multiple MCP servers
You can enable multiple MCP servers simultaneously. Your AI sees all tools from all enabled MCP servers.
Example setup:
- Marketing Suite (12 tools) → For campaigns
- Developer Tools (25 tools) → For code work
- Customer Support (8 tools) → For tickets
Total: 45 tools available in your AI, organized across 3 MCP servers.
Next Steps
- Understanding Providers - Learn about credentials and verification
- The Hub Endpoint - One URL for all your tools
- Composing MCP servers - Combine MCP servers into custom endpoints
- Quick Start Guide - Get your first MCP server working