> ## Documentation Index
> Fetch the complete documentation index at: https://docs.claude-space.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# MCPs Node

> Integrate external tools and services with Model Context Protocol

# MCPs Node (Model Context Protocol)

**MCPs** connect your agent to external tools, databases, and services.

***

## Overview

**Purpose:** Integrate external systems (databases, APIs, documentation)
**Connection:** Top-right handle of main agent card
**Color:** Blue
**Icon:** 🔌
**Required:** Optional (0 to many)

***

## What are MCPs?

Model Context Protocol (MCP) servers provide external tool access:

* Documentation lookup (Context7)
* Database connections (PostgreSQL, MongoDB)
* File systems
* Custom APIs

**Think:** Like plugins for your agent.

***

## MCP Structure

MCPs are configured with command + arguments:

```json theme={null}
{
  "context7": {
    "command": "npx",
    "args": ["-y", "@context7/mcp-server"],
    "env": {
      "CONTEXT7_API_KEY": "your-key"
    }
  }
}
```

Stored in `.mcp.json` at project root.

***

## Built-in MCPs

### Context7 (Documentation Lookup)

**Purpose:** Retrieve up-to-date library documentation

**Configuration:**

```json theme={null}
{
  "context7": {
    "command": "npx",
    "args": ["-y", "@context7/mcp-server"],
    "env": {
      "CONTEXT7_API_KEY": "your-api-key"
    }
  }
}
```

**Use case:** Agent looks up React/Vue/Python docs

***

### File System MCP

**Purpose:** Access local file system

**Configuration:**

```json theme={null}
{
  "filesystem": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-filesystem", "/allowed/path"]
  }
}
```

**Use case:** Read/write files outside project directory

***

### Database MCPs

**PostgreSQL:**

```json theme={null}
{
  "postgresql": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-postgres"],
    "env": {
      "DATABASE_URL": "postgresql://user:pass@localhost/db"
    }
  }
}
```

**Use case:** Query customer database, analytics

***

## Creating Custom MCPs

**Via Settings → Node Managers → MCPs:**

1. Click "Add New MCP"
2. Enter name (e.g., "Internal API")
3. Define command: `node server.js`
4. Add arguments: `["--port", "3000"]`
5. Set environment variables (API keys, URLs)
6. Save to workspace library

**MCP appears in sidebar → drag to canvas.**

***

## Configuration

**When clicked on canvas:**

* Edit command and args
* Set environment variables
* Configure connection settings
* Test connectivity

***

## Real-World Examples

### Example 1: Slack Integration

```json theme={null}
{
  "slack": {
    "command": "node",
    "args": ["slack-mcp-server.js"],
    "env": {
      "SLACK_TOKEN": "xoxb-your-token",
      "SLACK_CHANNEL": "#general"
    }
  }
}
```

**Agent can:**

* Send Slack messages
* Read channel history
* Search conversations

***

### Example 2: Redis Cache

```json theme={null}
{
  "redis": {
    "command": "node",
    "args": ["redis-mcp.js"],
    "env": {
      "REDIS_URL": "redis://localhost:6379"
    }
  }
}
```

**Agent can:**

* Cache frequently-used data
* Store session information
* Implement rate limiting

***

### Example 3: AWS Services

```json theme={null}
{
  "aws": {
    "command": "node",
    "args": ["aws-mcp-server.js"],
    "env": {
      "AWS_REGION": "us-east-1",
      "AWS_ACCESS_KEY_ID": "your-key",
      "AWS_SECRET_ACCESS_KEY": "your-secret"
    }
  }
}
```

**Agent can:**

* Upload files to S3
* Query DynamoDB
* Invoke Lambda functions

***

## Security Considerations

**⚠️ Important:**

* MCPs have full system access based on their permissions
* Store credentials in environment variables (not hardcoded)
* Use read-only database accounts when possible
* Audit MCP server code before use
* Restrict network access via firewalls

**Best practice:** Test MCPs in sandbox before production use.

***

## MCP vs. Skills

| Feature      | MCPs                           | Skills             |
| ------------ | ------------------------------ | ------------------ |
| **Purpose**  | External systems               | Agent capabilities |
| **Scope**    | System-wide                    | Project-specific   |
| **Language** | Any (runs as separate process) | Python/JS/shell    |
| **Example**  | Database connection            | Data processing    |

**Use both:** MCPs connect to external systems, skills process data.

***

## Troubleshooting

### MCP won't connect

**Check:**

1. Command path is correct
2. Dependencies installed (`npx` can find package)
3. Environment variables set correctly
4. Network connectivity (for remote services)
5. Authentication credentials valid

### MCP is slow

**Solutions:**

* Add caching layer
* Use connection pooling
* Implement rate limiting
* Choose geographically-close servers

***

## Next Steps

* [Hooks Node](/docs/node-types/hooks-node) - Lifecycle events
* [Sub-agents Node](/docs/node-types/subagents-node) - Specialized agents

***

<div align="center">
  [← Commands Node](/docs/node-types/commands-node) | [Agent Node](/docs/node-types/agent-node) | [Hooks Node →](/docs/node-types/hooks-node)
</div>