MCP Tools Reference

HardWire provides a set of built-in MCP (Model Context Protocol) tools that can be used in workflows. These tools enable AI agents to interact with the web, process text, and make HTTP requests.

Available Tools

web_search

Search the web for information using DuckDuckGo.

ID: tool_web_search Version: 1.0.0 Tags: search

Parameters

Parameter	Type	Required	Description
`query`	string	Yes	The search query
`limit`	number	No	Maximum number of results (1-10, default: 5)

Response

{
  "results": [
    {
      "title": "Result Title",
      "url": "https://example.com/page",
      "snippet": "A brief description of the result..."
    }
  ]
}

Example

{
  "method": "runTool",
  "params": {
    "id": "tool_web_search",
    "input": {
      "query": "latest AI news",
      "limit": 5
    }
  }
}

read_url

Fetch a URL and extract readable text content. Useful for reading web pages and extracting their main content.

ID: tool_read_url Version: 1.0.0 Tags: fetch

Parameters

Parameter	Type	Required	Description
`url`	string	Yes	The URL to fetch
`maxChars`	number	No	Maximum characters to return (500-20000, default: 8000)

Response

{
  "title": "Page Title",
  "text": "The extracted text content from the page..."
}

Example

{
  "method": "runTool",
  "params": {
    "id": "tool_read_url",
    "input": {
      "url": "https://example.com/article",
      "maxChars": 5000
    }
  }
}

Security Notes

Only HTTP and HTTPS protocols are allowed
Private IP addresses and localhost are blocked by default
An allowlist can be configured via the MCP_HTTP_REQUEST_ALLOWLIST environment variable

http_request

Make an HTTP API request. Supports all standard HTTP methods and custom headers.

ID: tool_http_request Version: 1.0.0 Tags: http

Parameters

Parameter	Type	Required	Description
`url`	string	Yes	The URL to request
`method`	string	Yes	HTTP method (GET, POST, PUT, DELETE, etc.)
`headers`	object	No	Custom request headers
`body`	any	No	Request body (ignored for GET/HEAD)
`timeoutMs`	number	No	Request timeout in milliseconds (1000-20000, default: 8000)

Response

{
  "status": 200,
  "headers": {
    "content-type": "application/json"
  },
  "body": "Response body content..."
}

Example

{
  "method": "runTool",
  "params": {
    "id": "tool_http_request",
    "input": {
      "url": "https://api.example.com/data",
      "method": "POST",
      "headers": {
        "Content-Type": "application/json",
        "Authorization": "Bearer token123"
      },
      "body": {
        "key": "value"
      }
    }
  }
}

Security Notes

Only HTTP and HTTPS protocols are allowed
Private IP addresses and localhost are blocked by default
Response body is limited to 20KB
Content-Type is automatically set to application/json for object bodies

summarize

Summarize a block of text into a shorter form. Extracts the first few sentences and truncates to a maximum word count.

ID: tool_summarize Version: 1.0.0 Tags: text

Parameters

Parameter	Type	Required	Description
`text`	string	Yes	The text to summarize
`maxWords`	number	No	Maximum words in summary (30-300, default: 120)

Response

{
  "summary": "A condensed version of the input text..."
}

Example

{
  "method": "runTool",
  "params": {
    "id": "tool_summarize",
    "input": {
      "text": "Long article text goes here...",
      "maxWords": 100
    }
  }
}

extract_json

Extract a JSON object or array from text. Useful for parsing structured data from unstructured responses.

ID: tool_extract_json Version: 1.0.0 Tags: json

Parameters

Parameter	Type	Required	Description
`text`	string	Yes	The text containing JSON
`fields`	string[]	No	Specific fields to extract (returns all if not specified)

Response

{
  "data": {
    "extracted": "json",
    "content": "here"
  }
}

Example

{
  "method": "runTool",
  "params": {
    "id": "tool_extract_json",
    "input": {
      "text": "The API returned: {\"status\": \"success\", \"count\": 42}",
      "fields": ["status", "count"]
    }
  }
}

Using Tools in Workflows

Tools are automatically available to AI agents during workflow execution. The agent can decide which tools to use based on the task at hand.

Tool Execution Flow

The AI agent analyzes the user's request
The agent selects appropriate tools to accomplish the task
Each tool is executed via the MCP Tools Server
Results are returned to the agent for further processing
The agent synthesizes results into a final response

Listing Available Tools

You can retrieve the list of available tools using the Get Tools endpoint:

curl -X GET "https://api.hardwire.dev/api/tools" \
  -H "Authorization: Bearer YOUR_ACCESS_TOKEN"

MCP Protocol

The MCP Tools Server implements the Model Context Protocol for standardized tool execution. The protocol supports four methods:

Method	Description
`listTools`	List all available tools with pagination
`searchTools`	Search tools by name, description, or tags
`getTool`	Get details of a specific tool
`runTool`	Execute a tool with given input

For more details on the MCP protocol, see the MCP Tools Server Overview.