IP Fabric MCP Server

Overview

The IP Fabric MCP Server is a Model Context Protocol (MCP) server that provides AI assistants with access to your IP Fabric network data. Key capabilities include:

Query network health using natural language
Analyze connectivity paths between endpoints with visual diagrams
Explore device inventory and configurations
Troubleshoot routing issues (BGP, OSPF, and more)
Validate network intents and compliance

The server exposes IP Fabric’s capabilities through standardized MCP tools, prompts, and resources that AI assistants can use to answer your network-related questions.

Prerequisites

HTTPS access to an instance of IP Fabric
API Token from your IP Fabric instance with appropriate permissions
MCP Server Terms & Conditions accepted and the server enabled in the IP Fabric UI (see Step 1 below)
An AI assistant that supports MCP — tested with Claude Desktop and VS Code with GitHub Copilot

Security Considerations

API Token: Your IP Fabric API token is used to authenticate AI assistant requests. The MCP server has no implicit access to IP Fabric data — every request is authorized solely through the provided API token. Generate a dedicated token with the minimum required permissions and rotate it regularly.
Network Access: The MCP server runs on the IP Fabric appliance and is accessed over HTTPS. Ensure appropriate network segmentation and firewall rules are in place between your AI client and the appliance.

Step 1 — Enable the MCP Server in IP Fabric

The MCP server is built into the IP Fabric appliance and can be enabled directly from the web UI.

Log in to your IP Fabric instance.
Navigate to Settings → Integration → MCP Server.
Read the Local MCP Terms of Use and check I have read and agree to MCP server T&C.
Toggle Enable MCP server to on.
Click Save.

Once enabled, the MCP server endpoint is available at https://<your-ipfabric-host>/mcp.

Step 2 — Generate an API Token

The MCP server authenticates to IP Fabric using an API token. The token’s Role-Based Access Control (RBAC) scope is fully respected — the MCP server (and therefore the AI assistant) can only access the data and actions permitted by the role assigned to the token.

Multiple AI clients can connect to the same MCP server simultaneously, each using its own API token. Every token is evaluated independently, so different users or teams can have separate access levels through the same MCP endpoint.

When creating the token, select a role with the minimum permissions required for your use case. For instructions on creating an API token, see API Tokens. For details on how the token is used for authentication, see Authentication — API Token.

Step 3 — Connect to an AI Assistant

Claude Desktop

At the time of writing, Claude Desktop only supports local (stdio) MCP servers. A local proxy is therefore needed to connect to the remote IP Fabric MCP server. This requires Node.js LTS to be installed on your machine.

1. Locate your Claude Desktop configuration:

Select your user profile, then Settings, then Developer. Clicking Edit will open claude_desktop_config.json.

2. Add the MCP server configuration:

{
  "mcpServers": {
    "ipfabric-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://<your-ipfabric-host>/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer <YOUR_API_TOKEN_HERE>"
      }
    }
  }
}

Note

We use the npx command to download and start a proxy called mcp-remote in Node.

npx looks up an npm package and runs its executable, downloading it on the fly.
-y auto-confirms the install prompt.
mcp-remote is the name of the npm package being executed.
--header passes the Authorization header on every request to the MCP server.
The token value is kept in env to avoid exposing it directly in the args. Use the format Bearer <token>.

3. Restart Claude Desktop to apply the configuration.

4. Verify the connection: Look for the MCP tools icon in Claude’s interface.

VS Code with GitHub Copilot

1. Locate your VS Code MCP configuration:

macOS: ~/Library/Application Support/Code/User/mcp.json
Linux: ~/.config/Code/User/mcp.json
Windows: %APPDATA%\Code\User\mcp.json

2. Add the server configuration:

{
  "servers": {
    "ipfabric-mcp": {
      "url": "https://<your-ipfabric-host>/mcp",
      "type": "http",
      "headers": {
        "Authorization": "Bearer <YOUR_API_TOKEN_HERE>"
      }
    }
  }
}

3. Start the MCP server from VS Code:

Open Command Palette: ⌘+Shift+P (macOS) or Ctrl+Shift+P (Windows/Linux)
Run: MCP: List Servers
Select: ipfabric-mcp → Start Server

4. Verify in the Output panel (View → Output) that tools are loaded successfully.

Note

If you encounter JSON validation errors, disable JSON validation in VS Code settings: search for JSON > Validate: Enable and uncheck it.

Step 4 — Use the MCP Server

Available Tools

The MCP server provides specialized tools for network analysis. Your AI assistant automatically selects the appropriate tool based on your prompt.

Network Health Assessment

Tool: ipf_network_health_assess

Provides a comprehensive overview of your network health including:

Snapshot freshness — How recent is the network discovery data
Intent verification — Built-in and custom intent rule results
Inventory issues — Device health, uptime, licensing
Routing stability — BGP/OSPF neighbor states
Path check results — End-to-end connectivity validation

Example prompts:

Check network health
Show me critical issues only
What's the BGP status across my network?
Are there any routing problems?

Path Lookup Tools

Unicast Path Lookup

Tool: ipf_pathlookup_unicast

Traces the path between two IP addresses and returns structured path data.

Example prompts:

Show me the path from 10.0.1.5 to 10.0.2.10
Why can't 192.168.1.100 reach 172.16.0.50?

Parameters:

Parameter	Required	Description
`src`	Yes	Source IP address
`dst`	Yes	Destination IP address
`snapshotId`	Yes	UUID of the snapshot to analyze
`groupBy`	Yes	Grouping method: `siteName`, `routingDomain`, or `stpDomain`
`protocol`	No	Protocol: `icmp` (default), `tcp`, or `udp`
`l4Options`	No	Layer 4 options (ports, flags)
`vrf`	No	VRF name if applicable
`includeGraphLayout`	No	Include graph layout data in the response. By default, layout data is stripped from the result; set to `true` to include it explicitly. Default: `false`

Host-to-Gateway Path Lookup

Tool: ipf_pathlookup_host-to-gateway

Traces the path from a host to its default gateway and returns structured path data.

Example prompts:

Show path from 192.168.1.100 to its gateway
How does host 10.0.5.25 reach its gateway?

Parameters:

Parameter	Required	Description
`src`	Yes	Source IP address
`snapshotId`	Yes	UUID of the snapshot to analyze
`groupBy`	Yes	Grouping method: `siteName`, `routingDomain`, or `stpDomain`
`vrf`	No	VRF name if applicable
`includeGraphLayout`	No	Include graph layout data in the response. By default, layout data is stripped from the result; set to `true` to include it explicitly. Default: `false`

Multicast Path Lookup

Tool: ipf_pathlookup_multicast

Traces multicast traffic paths from source to receivers and returns structured path data.

Example prompts:

Show multicast path for group 239.1.1.1 from source 10.0.1.5
Analyze multicast distribution for 239.255.0.1

Parameters:

Parameter	Required	Description
`src`	Yes	Source IP address
`group`	Yes	Multicast group address
`receiver`	Yes	Comma-separated receiver IP addresses
`snapshotId`	Yes	UUID of the snapshot to analyze
`groupBy`	Yes	Grouping method
`includeGraphLayout`	No	Include graph layout data in the response. By default, layout data is stripped from the result; set to `true` to include it explicitly. Default: `false`

Path Diagram Tools

Each path lookup tool has a companion that returns a PNG diagram of the path instead of structured data.

Diagram Tool	Corresponding Path Lookup
`ipf_png_pathlookup_unicast`	`ipf_pathlookup_unicast`
`ipf_png_pathlookup_host-to-gateway`	`ipf_pathlookup_host-to-gateway`
`ipf_png_pathlookup_multicast`	`ipf_pathlookup_multicast`

The diagram tools accept the same parameters as their corresponding path lookup tools (excluding includeGraphLayout) and return a visual network path image.

Example prompts:

Trace route from server A to server B with diagram
Show multicast path for group 239.1.1.1 from source 10.0.1.5 with diagram

API Discovery Tools

For advanced queries not covered by specialized tools, use the API discovery workflow.

API Endpoint Search

Tool: ipf_api_endpoint_search

Finds relevant API endpoints using natural language.

Example prompts:

Find endpoints for device inventory
Search for BGP neighbor API

API Endpoint Details

Tool: ipf_api_endpoint_details

Gets detailed information about a specific endpoint including parameters and response schema.

API Invoke

Tool: api_invoke

Executes an API call with the specified parameters.

Example workflow:

Search: "device inventory endpoints"
Get details for the relevant endpoint
Invoke the endpoint with appropriate filters

Available Prompts

Prompts provide structured templates for complex analysis tasks.

Path Analysis Prompt

Name: path_analysis

A guided prompt for detailed path analysis between network points.

How to use:

Claude Desktop — Type / in the chat and select path_analysis.
VS Code with GitHub Copilot — Type @ipfabric-mcp in the chat to invoke the MCP server, then describe the path analysis you need.

Parameters:

snapshot — Snapshot name to analyze
source — Source IP or device name
destination — Destination IP (for unicast)
group — Multicast group (if applicable)
protocol — TCP, UDP, or ICMP (optional)

This prompt generates a comprehensive analysis including hop-by-hop breakdown, issue identification, and topology improvement recommendations.

Available Playbooks (Resources)

Playbooks are reference guides provided by the MCP server that help the AI assistant make better decisions. They contain domain-specific knowledge about network troubleshooting workflows and best practices. How playbooks are selected or attached depends on the AI client you are using — in Claude Desktop, they appear as resources in the MCP tools panel; in VS Code, they are available as context for the AI assistant.

Usage Examples

Health Monitoring

Basic health check:

Check network health

Focus on specific areas:

Show me BGP issues only
Check routing protocol status with critical issues only

Investigate specific problems:

Are there any devices with high CPU utilization?

Path Analysis

Basic connectivity:

Show the path from 10.0.1.5 to 10.0.2.10 with diagram

With protocol specification:

Trace the path from 192.168.1.10 to 172.16.0.5 using TCP port 443

Host to gateway:

Show how 192.168.1.100 reaches its default gateway

VRF-aware path:

Show path from 10.1.1.1 to 10.2.2.2 in VRF MGMT

Multicast:

Show multicast path for group 239.1.1.1 from source 10.0.1.5 with diagram

Device Queries

Site-based inventory:

Show all Cisco devices in site HQ

Status queries:

Find devices with uptime less than 1 day
List devices with telnet enabled

Protocol Analysis

BGP:

Show BGP neighbors that are not in Established state

OSPF:

List OSPF neighbors not in Full state

General routing:

Are there any routing adjacency issues?

Troubleshooting

Connection Issues

Cannot reach the MCP server:

Verify the MCP server is enabled in Settings → Integration → MCP Server.
Ensure the API token is valid and has not been revoked.

API Errors

401 Unauthorized:

Verify your API token is correct.
Check that the token has not been revoked.
Ensure the token has sufficient permissions.

404 Not Found:

Verify the MCP server is enabled and that the URL is correct.

Snapshot Errors

“Snapshot not found”:

Ensure snapshots are loaded in IP Fabric.
The AI will automatically list available snapshots if needed.
You can specify $last to use the most recent snapshot.

Common Questions

How do I disable the MCP server?

Navigate to Settings → Integration → MCP Server, toggle Enable MCP server to off, and click Save.

Can I use a different API token for MCP?

Yes. Generate a dedicated token under Settings → Integration → API Tokens with the minimum required permissions and configure your AI client to use it.