ON THIS PAGE
 

Using the Mist MCP Server

You can connect to the Mist MCP server and then use conversational English to retrieve performance statistics, usage details, and other such information about your organizations in the Mist cloud.

Security Considerations for Using AI

Warning:

The Juniper Mist Model Context Protocol (MCP)server gives AI assistants access to your network configuration and operational data, which may include sensitive information such as preshared keys (PSKs), RADIUS secrets, and SNMP credentials.

Query results are returned to your MCP client and become part of the conversation context sent to whichever AI model that client uses. Before connecting the MCP server to any AI assistant, keep the following in mind:

  • Know where your data goes. Some cloud-based AI services may use conversation data — including MCP responses — to train future models. Know your organization's data-handling policy before connecting production orgs, and review your AI provider's data-retention terms. Only use AI services that have been reviewed and approved by your organization's security and compliance teams.

  • Consider self-hosted AI. To keep sensitive network data from leaving your environment, consider locally hosted LLM solutions (for example, Ollama, LM Studio) instead of cloud-based services.

  • Apply least privilege. Use API tokens with the minimum permissions required. Prefer read-only tokens, and avoid granting broader access than the task demands.

  • Protect your configuration file. The Claude Desktop configuration file stores your API token in plaintext. Secure it with appropriate file permissions and never share it or commit it to version control.

Natural Language Queries Using AI

Wouldn’t it be nice to be able to leverage the power of an LLM in the context of your Mist organization? Then you could use conversational English to ask the LLM questions about your Mist organization and get back dynamic answers about the configuration, SLE statistics, inventory, and significant events.

That’s what the Juniper Mist MCP server allows you to do. It creates a secure, purpose-built data bridge between your AI assistant and the AI mode so your MCP client can access your organizations on the Mist cloud. The MCP server exposes tools (listed in Table 1) to the LLM, that the LLM can use to pull read-only information from your org. Your MCP client, like Claude or ChatGPT, consolidates the data from the Mist cloud and provides you with the information in conversational form.

Thus, you can use Mist MCP server to ask a question such as,

“Did yesterday’s firmware upgrade cause any Wi-Fi disruptions at the Washington DC site?”

and get back an answer like,

“No. I just checked the logs for the DC sites and there are no reports of disruption. This is also collaborated by the fact that performance statistics for the past 24 hours are consistent with historical trends for the Washington DC site.”

Additional examples are shown below.

Troubleshooting

  • Which NAC rules are matching laptop "West-WIN-Laptop", and why is it landing on the wrong VLAN?
  • Client aa:bb:cc:dd:ee:ff can't get on the network — what's happening?
  • Show me every DHCP failure at the London site in the last four hours.

Health and monitoring

  • Are there any network issues right now?
  • Which five sites have the worst coverage SLE this week?
  • Are any BGP peers down across the org?

Configuration and audit

  • List all sites that are running AP firmware 0.14 or earlier.
  • List every WLAN with the guest portal enabled.
  • Show me any differences in the two campus switch templates.

See Using the Juniper Mist MCP Server with Claude Desktop for instruction on how to set up your MCP client to work with the Mist MCP server.

Mist MCP Tools

In order to dynamically answer questions in real-time about a given Mist organization, the Mist MCP server exposes a rich set of tools based on the read-only public Mist APIs. The tools span the entire Mist ecosystem.

Note that the Mist MCP server is not intended to be used for unattended automation, closed-loop remediation, or any operations designed to create or modify settings.

  • Site and org structure

  • Service level expectations

  • Marvis actions

  • Device inventory

  • Wireless, wired, WAN, and NAC clients

  • WLAN, switch, and WAN configurations

  • NAC rules and policies

  • Guest authorizations

  • Events, including device, client, and rogue events

  • Alarms

  • Routing and tunnel states

  • Live status

The table below describes the seven tools exposed by the MCP server. Each tool is capable of leveraging multiple API calls.

Table 1: MCP Data Tools
Tool Answers Data Nature
get_mist_self Who am I, and which orgs can I see? Identity
get_mist_config How is the network configured? Intended state
get_mist_stats What is the network doing right now? Current-state snapshot
search_mist_data What happened, and to whom? Historical events and records
get_mist_insights What's wrong? How good is the experience? Marvis analytics and SLE
find_mist_entity Where is this MAC, IP, or hostname? Universal lookup
get_mist_constants What are the valid values for X? Reference data

AI Technologies

There are multiple AI components involved in the chain of events needed to go from asking an LLM at your laptop to getting an answer from the Mist cloud. To help sort it all out, the main AI technologies are described below to provide some useful background information, as well as an example workflow.

  • The AI model, or LLM — This is the so-called “brain-in-a-jar”, that is, the reasoning engine that has been trained on trillions of examples from the Internet. Examples include Claude (Anthropic), GPT-5.6 (OpenAI), and self-hosted models such as LLaMA through Ollama or LM Studio. The AI model understands your question, determines which MCP tools are needed to get the answer, and then, with the data from Mist cloud, provides the facts of your organization in conversational form.

  • The AI Application — This is the MCP-capable assistant, that is, the software you interact with to chat. Examples include Claude Desktop, ChatGPT, and CoPilot in VS Code. It includes the chat window, but actually coordinates the entire loop with the AI model.

  • The MCP Client — This is the protocol translator, which is part of the AI application. It connects to the MCP server and formats response/requests according to the MCP protocol. It also retrieves the list of tools and tool information so the AI model knows what it can invoke.

    • In the Mist MCP configuration, the client connects to the remote MCP server at https://mcp.ai.juniper.net/mcp/mist, authenticating with your Mist API token through an Authorization: Bearer header.

  • The Juniper Mist MCP Server — This is the data bridge between the AI model and the Mist cloud. It is hosted on the HPE cloud and exposes a curated set of tools that are backed by one or more Mist REST API calls. The MCP server determines which Mist REST API endpoints to query, constructs the requests with appropriate parameters (site ID, time range, event type). It authenticates using the security token for the given Mist organization, and executes the API call. By design, the Juniper MCP server only supports read-only operations. It does not create, modify, or delete any configuration in your Mist organization.

  • The Mist Cloud — This is the location of your Mist organization data.

A typical work flow is described here:

A User engages with an AI assistant (such as VS Code, Claude Desktop, or Gemini) to ask a question about their Mist organization. The AI assistant hosts an MCP client, which connects to the MCP server. The MCP server provides the client with a list of tools and resources that are available. The MCP client sends this list, along with the user’s question, to an AI model (also called an LLM). The LLM, (such as Claude 3.5 or GPT-5.6), reasons through the question, and then determines which tools can best provide the raw data needed to answer the user’s question. It informs the MCP client, which invokes the tool, and the MCP server calls the Mist API to handle authentication into the org, and get the requested data. The MCP server returns the data to the MCP client, which passes the data back to the AI model. The AI Model takes the raw data and synthesizes it into a conversational answer to the User’s question.

Common Use Cases for Leveraging an MCP Server

Use the Mist MCP server for the following use cases:

  • Network operations and helpdesk—You can automate the first response on help tickets that arrive as a MAC address and a complaint; the investigation path from identifier to root cause supports fast and direct case resolution.

  • Network engineers and architects—The AI assistant is especially convenient for asking a series of exploratory, one-off questions; likewise, it is an excellent way to conduct ad hoc audits and reporting across sites because you don't need to write throwaway scripts.

  • Developers and integrators—The AI assistant also excels as a way to explore the data model interactively, for example to spot-check and understand a planned integration or prototype.