MCP Client Tool Node

The MCP Client Tool node connects to an MCP (Model Context Protocol) server and exposes its tools to AI Agents. Unlike built-in Tool nodes, MCP Client Tool dynamically imports remote tools at runtime — the AI Agent can call any tool registered on the connected MCP server, enabling integration with a vast ecosystem of external services and custom tooling.

Prerequisites: You must have at least one MCP server configured in your workspace. If no servers are available, you'll be prompted to add one via the MCP Server Management page.

Core Concepts

What is MCP?

The Model Context Protocol is an open standard that enables AI applications to connect with external tools and data sources through a unified interface. An MCP server can expose anything from database queries to API integrations as callable "tools" that AI Agents can discover and invoke.

How the MCP Client Tool Works

MCP Server (external)                    Your Workflow
┌─────────────────────┐                  ┌──────────────────────┐
│  Tools registered:  │   JSON-RPC       │  Chat Trigger        │
│  - get_weather      │ ←────────────── →│    → AI Agent Node   │
│  - search_docs      │                  │        Tools:        │
│  - create_ticket    │                  │          MCP Client  │
│  - query_database   │                  │            Tool      │
└─────────────────────┘                  │    → Answer Node     │
                                         └──────────────────────┘

The MCP Client Tool connects to the MCP server at runtime
Available tools are discovered and registered with the AI Agent
The AI Agent calls the appropriate MCP tool based on user needs
Results are returned to the AI Agent for synthesis

Use Cases

Typical Applications

External API Integration — Connect to third-party services via their MCP servers (e.g., weather, maps, email)
Database Access — Let AI Agents query databases through MCP database servers
Custom Internal Tools — Expose company-specific tools and data sources
File System Operations — Allow AI Agents to read and manage files via filesystem MCP servers
Cross-Platform Orchestration — Bridge multiple services through different MCP servers in one workflow

Node Configuration

Setup Workflow

Configuring the MCP Client Tool involves four steps:

Select an MCP server — Pick from installed servers or add a new one
Choose which tools to include — All tools, specific tools, or all except specific ones
Select tools (if applicable) — Pick individual tools from the server's catalog
Set timeout — Configure how long to wait for MCP tool responses

Basic Settings

MCP Server (mcpServerId)

Select the MCP server whose tools you want to expose to the AI Agent.

Field Properties:

Required field
Dropdown populated from installed MCP servers in your workspace
If no servers are available, the dropdown is empty

If no MCP servers are available:

A "Manage MCP Servers" link button appears below the dropdown. Click it to open the MCP Server Management page where you can:

Install community MCP servers from the catalog
Create custom MCP servers with your own endpoints
Configure authentication credentials for MCP servers

The page opens in a new tab, so you can configure servers without leaving your workflow.

After selecting a server: The available tools are fetched from the server. The Include Tools setting becomes active.

Include Tools (include)

Control which tools from the MCP server are exposed to the AI Agent.

Available Options:

Option	EN Label	CN Label	Description
`all`	All Tools	全部工具	Expose all tools from the MCP server (default)
`selected`	Selected Tools	选定工具	Only expose specifically chosen tools
`except`	All Except Selected	全部（排除选定）	Expose all tools except those excluded

Recommendations:

yaml

Simple servers (≤10 tools): Use "All Tools"
  → AI Agent has full access, simpler configuration

Many tools or sensitive ones: Use "Selected Tools" or "All Except Selected"
  → Fine-grained control over which tools are available
  → Prevents AI from accidentally calling inappropriate tools

Example:
  Server with 50+ tools → "Selected Tools" → pick 5-8 relevant ones
  Server with 10 tools but 2 are admin-only → "All Except Selected" → exclude admin tools

Tools (tools)

When Include Tools is set to selected or except, a tool picker appears:

Selected Tools mode: Check the tools you want the AI Agent to be able to call (required — at least one must be selected)
All Except Selected mode: Check the tools you want to exclude from the AI Agent

Tool names and descriptions from the MCP server are displayed to help you choose.

Timeout (timeout)

Maximum time in milliseconds to wait for a response from the MCP server when a tool is called.

Field Properties:

Number type (positive integer)
Default value: 60000 (60 seconds)
Supports expressions

Configuration Suggestions:

yaml

Fast tools (simple queries):     10000-30000  (10-30s)
Medium tools (API calls):        30000-60000  (30-60s, default)
Slow tools (complex processing): 60000-120000 (60-120s)

Note: If the MCP server doesn't respond within the timeout, the tool call fails and the error is returned to the AI Agent.

Advanced Settings

Node Description (nodeDescription)

Add a custom description for documentation purposes:

yaml

nodeDescription: "Connects to the Weather MCP server. Provides get_weather and get_forecast tools."

How It Works

Runtime Flow

1. Workflow reaches MCP Client Tool node
     ↓
2. Node connects to MCP server via JSON-RPC
     ↓
3. MCP session initialized (handshake + protocol negotiation)
     ↓
4. Available tools fetched from server
     ↓
5. Tools filtered based on "Include Tools" setting
     ↓
6. Filtered tools registered with parent AI Agent
     ↓
7. AI Agent can now call these tools by name
     ↓
8. When called: request forwarded to MCP server → response returned to AI

Tool Naming Convention

Tools from MCP servers are registered with a composite name:

{nodeId}--{toolName}

For example, if your MCP Client Tool node's ID is abc123 and the server has a tool named get_weather, the AI Agent calls it as abc123--get_weather. This prevents naming conflicts between multiple MCP Client Tool nodes.

You don't need to worry about this — the AI Agent handles tool naming automatically.

Error Handling

If the MCP server is unreachable or a tool call fails:

Server not found — Returns an error: "MCP server not found"
Request timeout — Returns an error if the server doesn't respond within the configured timeout
Tool call error — If the MCP tool itself returns an error, the error content is returned to the AI Agent
Invalid credential — Returns an error if the server requires authentication with an unsupported credential type

The AI Agent typically handles these gracefully — it will try alternative approaches or inform the user of the limitation.

Workflow Examples

Example 1: Weather-Aware AI Agent

Chat Trigger
  → AI Agent Node
    System Prompt: "You are a helpful assistant with access to weather data."
    User Prompt: $('Chat Trigger').message

    Tools:
      └─ MCP Client Tool (Weather Server)
           mcpServerId: "weather-mcp-server"
           include: "all"
           timeout: 30000

  → Answer Node

User: "What's the weather like in Tokyo?"
AI Agent:
  1. Calls MCP tool: abc123--get_weather({ city: "Tokyo" })
  2. MCP server returns: { temperature: 22, condition: "sunny", humidity: 55 }
  3. AI: "Tokyo is sunny and 22°C with 55% humidity today."

Example 2: Multi-Tool MCP with Selective Access

Chat Trigger
  → AI Agent Node
    System Prompt: "You are a customer support agent. Use available tools to help customers."
    User Prompt: $('Chat Trigger').message

    Tools:
      └─ MCP Client Tool (Internal Tools Server)
           mcpServerId: "internal-tools-mcp"
           include: "selected"
           tools: ["query_customer", "lookup_order", "check_inventory"]
           timeout: 60000

      Note: The server has 20 tools, but only 3 are selected.
            Tools like "delete_order" and "modify_pricing" are not exposed.

  → Answer Node

User: "What's the status of my order #12345?"
AI Agent:
  1. Calls MCP tool: lookup_order({ orderId: "12345" })
  2. Returns order status
  3. AI: "Your order #12345 is shipped and arriving tomorrow."

Example 3: Multiple MCP Servers

Chat Trigger
  → AI Agent Node
    System Prompt: "You have access to multiple service platforms. Use the right tool for each task."
    User Prompt: $('Chat Trigger').message

    Tools:
      ├─ MCP Client Tool (CRM Server)
      │    mcpServerId: "salesforce-mcp"
      │    include: "selected"
      │    tools: ["search_contacts", "get_opportunity"]
      │    timeout: 60000
      │
      ├─ MCP Client Tool (Email Server)
      │    mcpServerId: "gmail-mcp"
      │    include: "all"
      │    timeout: 30000
      │
      └─ MCP Client Tool (Filesystem Server)
           mcpServerId: "filesystem-mcp"
           include: "all"
           timeout: 15000

  → Answer Node

User: "Find John Doe in CRM, draft an email for him, and save it as a draft"
AI Agent:
  1. Calls CRM tool: search_contacts({ name: "John Doe" }) → Finds contact
  2. Calls Email tool: create_draft({ to: "john@example.com", ... }) → Draft created
  3. AI: "I found John Doe (john@example.com) and saved an email draft."

Example 4: Secure Data Access with Filtering

Chat Trigger
  → AI Agent Node
    System Prompt: "You are a data analyst. Query the database but never modify or delete data."
    User Prompt: $('Chat Trigger').message

    Tools:
      └─ MCP Client Tool (Database Server)
           mcpServerId: "postgres-mcp"
           include: "except"
           tools: ["delete_records", "drop_table", "truncate_table", "alter_schema"]
           timeout: 60000

      Note: Using "All Except Selected" to exclude dangerous admin tools.
            The AI can query and read data, but cannot modify or delete.

  → Answer Node

User: "Show me the top 10 customers by revenue this quarter"
AI Agent:
  1. Calls MCP tool: run_query({ sql: "SELECT ... FROM customers ... ORDER BY revenue DESC LIMIT 10" })
  2. Returns formatted results

Best Practices

1. Use Selected Mode for Production

In production workflows, avoid all mode. Explicitly select which tools the AI Agent can access:

javascript

// Good: Explicit selection
include: "selected"
tools: ["get_weather", "get_forecast", "get_alerts"]

// Risky: Full access in production
include: "all"
// AI might accidentally call admin or destructive tools

2. Set Appropriate Timeouts

Match the timeout to your MCP server's typical response time:

javascript

// Local/lightweight servers
timeout: 15000  // 15 seconds

// External API servers
timeout: 30000  // 30 seconds

// Complex processing servers
timeout: 60000  // 60 seconds (default)

// Long-running operations
timeout: 120000  // 2 minutes

3. Pre-install MCP Servers Before Building Workflows

Configure your MCP servers in the MCP Server Management page before building your workflow:

Go to MCP Servers in the workspace settings
Add the servers you need (from catalog or custom)
Test the connection
Then use them in MCP Client Tool nodes

4. Use Multiple MCP Client Tool Nodes for Different Purposes

If your AI Agent needs tools from multiple MCP servers, use separate MCP Client Tool nodes — one per server. This gives you:

Independent tool selection per server
Different timeouts per server
Clearer configuration and debugging

5. Test MCP Tools Before Production

Test each MCP tool individually before exposing it to the AI Agent:

Register the tool with an AI Agent in a test workflow
Ask the AI to call the tool with test inputs
Verify the response format and timing
Adjust tool selection and timeout as needed

6. Monitor MCP Tool Usage

Use workflow execution logs to track MCP tool calls:

Which tools are being called most frequently?
Are any tools consistently timing out?
Are there tools the AI never uses (candidates for removal)?
What errors are occurring?

FAQ

Q1: What's the difference between MCP Client Tool and HTTP Request Tool?

Aspect	HTTP Request Tool	MCP Client Tool
Protocol	Direct HTTP REST calls	JSON-RPC via MCP protocol
Tool Discovery	Manual configuration	Automatic from MCP server
Multi-tool	One tool = one endpoint	One node = many tools
Standardization	Custom per API	Standard MCP tool schema
Best For	Simple API calls	Complex tool servers with many endpoints

Q2: How do I add an MCP server?

A: Click the "Manage MCP Servers" button in the MCP Client Tool node, or navigate to the MCP Servers page from workspace settings. From there you can:

Browse catalog — Install community MCP servers (e.g., filesystem, GitHub, Slack)
Create custom — Add your own MCP server by providing a URL and optional credentials
Configure credentials — Set up authentication for servers that require it

Q3: Can I connect to multiple MCP servers in one workflow?

A: Yes. Add one MCP Client Tool node per server. Each node connects to one MCP server and exposes its tools. The AI Agent can call tools from any connected server.

Q4: What happens if the MCP server is unreachable?

A: If the MCP server cannot be reached at runtime, the tool call fails with an error message. The AI Agent typically handles this gracefully — it will try alternative approaches or tell the user the service is temporarily unavailable. Check your On Error settings on the parent AI Agent node.

Q5: Does the AI Agent know which tools are available?

A: Yes. When the MCP Client Tool registers with the AI Agent, all tools (filtered by your include/tools settings) are added to the AI's available tool list with their names and descriptions. The AI uses this information to decide which tool to call for each user request.

Q6: Can I use expressions in the timeout field?

A: Yes. The timeout field supports expressions, allowing dynamic timeout values:

javascript

// Dynamic timeout based on workflow context
$('Config').mcpTimeout || 60000

// Conditional timeout
$('Chat Trigger').isPriority ? 15000 : 60000

Next Steps

AI Agent Node — Learn about the main AI Agent node
HTTP Request Tool Node — Add direct API calling capability
Code Tool Node — Add custom code tools

MCP Server Management — Install and configure MCP servers
Think Tool Node — Add structured reasoning for AI Agents
Sub-workflow Tool Node — Register sub-workflows as tools

Chat Trigger

MCP Client Tool Node ​

Core Concepts ​

What is MCP? ​

How the MCP Client Tool Works ​

Use Cases ​

Typical Applications ​

Node Configuration ​

Setup Workflow ​

Basic Settings ​

MCP Server (mcpServerId) ​

Include Tools (include) ​

Tools (tools) ​

Timeout (timeout) ​

Advanced Settings ​

Node Description (nodeDescription) ​

How It Works ​

Runtime Flow ​

Tool Naming Convention ​

Error Handling ​

Workflow Examples ​

Example 1: Weather-Aware AI Agent ​

Example 2: Multi-Tool MCP with Selective Access ​

Example 3: Multiple MCP Servers ​

Example 4: Secure Data Access with Filtering ​

Best Practices ​

1. Use Selected Mode for Production ​

2. Set Appropriate Timeouts ​

3. Pre-install MCP Servers Before Building Workflows ​

4. Use Multiple MCP Client Tool Nodes for Different Purposes ​

5. Test MCP Tools Before Production ​

6. Monitor MCP Tool Usage ​

FAQ ​

Q1: What's the difference between MCP Client Tool and HTTP Request Tool? ​

Q2: How do I add an MCP server? ​

Q3: Can I connect to multiple MCP servers in one workflow? ​

Q4: What happens if the MCP server is unreachable? ​

Q5: Does the AI Agent know which tools are available? ​

Q6: Can I use expressions in the timeout field? ​

Next Steps ​

Related Resources ​

MCP Client Tool Node

Core Concepts

What is MCP?

How the MCP Client Tool Works

Use Cases

Typical Applications

Node Configuration

Setup Workflow

Basic Settings

MCP Server (mcpServerId)

Include Tools (include)

Tools (tools)

Timeout (timeout)

Advanced Settings

Node Description (nodeDescription)

How It Works

Runtime Flow

Tool Naming Convention

Error Handling

Workflow Examples

Example 1: Weather-Aware AI Agent

Example 2: Multi-Tool MCP with Selective Access

Example 3: Multiple MCP Servers

Example 4: Secure Data Access with Filtering

Best Practices

1. Use Selected Mode for Production

2. Set Appropriate Timeouts

3. Pre-install MCP Servers Before Building Workflows

4. Use Multiple MCP Client Tool Nodes for Different Purposes

5. Test MCP Tools Before Production

6. Monitor MCP Tool Usage

FAQ

Q1: What's the difference between MCP Client Tool and HTTP Request Tool?

Q2: How do I add an MCP server?

Q3: Can I connect to multiple MCP servers in one workflow?

Q4: What happens if the MCP server is unreachable?

Q5: Does the AI Agent know which tools are available?

Q6: Can I use expressions in the timeout field?

Next Steps

Related Resources