Installing LLM Brand Monitor MCP

This guide walks you through connecting the LLM Brand Monitor (LBM) MCP server to popular AI assistants and developer tools. Once connected, you can manage brand-monitoring projects, run scans across 350+ LLMs, and analyze results with natural-language commands.

Table of contents

• Introduction
• Which version should I use — HTTP or stdio?
• Getting your credentials
• HTTP (remote) installation
  • ChatGPT
  • Claude Web (claude.ai)
• stdio (local) installation
  • Claude Desktop (Windows & macOS)
  • Claude Code
  • Codex CLI
  • Gemini CLI
  • OpenClaw
• Verifying the connection
• Troubleshooting

Introduction

The LLM Brand Monitor MCP server exposes 17 tools that let any Model Context Protocol–compatible client talk to the LBM platform. You can ask the assistant to list projects, start scans, fetch transcripts, compare competitor mentions, and check your credit balance — all in plain English (or any other language the model speaks).

The server ships in two flavors:

Which version should I use — HTTP or stdio?

Use HTTP when:

• Your client is a hosted web app (ChatGPT, Claude Web) that has no way to launch a local process.
• You don’t want to install Node.js locally or paste your API key into a config file.
• You want browser-based sign-in with the same Google/email account you already use on llmbrandmonitor.com.
• You want zero-maintenance updates — the server is upgraded centrally.

Use stdio when:

• Your client is a desktop or terminal app that spawns subprocesses (Claude Desktop, Claude Code, Codex, Gemini CLI, OpenClaw, Cursor, Windsurf).
• You prefer a long-lived API key over OAuth flows.
• You’re working offline-friendly or behind a corporate proxy where browser-based OAuth is awkward.
• You want to pin a specific MCP server version, or run a fork/local build.

If you’re still not sure: HTTP is the lowest-friction path for end users; stdio gives you more control and is the only option for developer tools.

Getting your credentials

Both versions start from the same place — an account on llmbrandmonitor.com with an active billing setup.

[!IMPORTANT] MCP access requires a paid subscription or Pay-as-you-go. The free tier does not include API/MCP access. Pick a plan at https://llmbrandmonitor.com/pricing before continuing — on a free plan every tool call will fail with UNAUTHORIZED (the API key is rejected because the plan isn’t whitelisted), and once a paid plan runs out of credits calls fail with INSUFFICIENT_CREDITS.

1. Sign up or log in at https://llmbrandmonitor.com.
2. Activate a subscription plan or enable Pay-as-you-go at https://llmbrandmonitor.com/pricing.
3. Open your profile page.
4. For stdio: copy your API key (it starts with lbm_).
5. For HTTP: no key is needed — you’ll authorize the connection in the browser the first time you use it.

HTTP (remote) installation

The hosted server lives at:

You don’t need to install or run anything locally — just paste this URL into your client and complete OAuth.

ChatGPT

Available in ChatGPT plans that support MCP Apps (Plus, Pro, Business, Enterprise). The exact menu names may shift slightly as OpenAI iterates on the UI.

1. Open ChatGPT and click your avatar → Settings.

2. Go to Apps (in some plans you may need to enable developer/MCP apps under Beta features first), then click Create app in the top-right corner.

   

   In Settings → Apps you’ll see any apps you’ve already enabled. Click Create app to add LBM.

3. Fill in the New App form:

   • Name: LLM Brand Monitor
   • Description (optional): Track brand visibility in LLMs
   • MCP Server URL: https://mcp.llmbrandmonitor.com/mcp
   • Authentication: OAuth
   • Leave Client ID / Secret empty — the server uses Dynamic Client Registration, so ChatGPT will register itself automatically.
   • Tick I understand and want to continue under the “Custom MCP servers introduce risk” warning, then click Create.

   

   Filled-in form. The risk warning is standard for any custom MCP server — ChatGPT shows it for every third-party app.

4. ChatGPT opens a browser tab on llmbrandmonitor.com. Sign in if you’re not already logged in.

5. On the consent screen, review the requested scopes (projects:read, projects:write, scans:read, scans:write, results:read, usage:read, offline_access) and click Allow.

   

   The consent page is hosted on llmbrandmonitor.com. Tick Remember my decision if you don’t want to re-authorize on every new ChatGPT session.

6. You’re redirected back to ChatGPT — the app now shows as Connected. Start a new chat and enable the app for that conversation: click the + in the composer → More → tick LLM Brand Monitor.

   

   Once ticked, the app name appears next to the + icon and stays active for the rest of that conversation.

7. Send a prompt — ChatGPT will call the LBM tools and answer with live data from your account.

   

   Example output: ChatGPT lists your projects and the top competitors LLMs mention for each one. The “Called tool” chips show every MCP call ChatGPT made — click them to see raw tool input/output.

Claude Web (claude.ai)

1. Open https://claude.ai and sign in.

2. Click your avatar → Settings → Connectors (in some regions the section is labeled Integrations). Scroll to the bottom of the list and click Add custom connector.

   

   The Connectors page lists everything Claude can already talk to. The Add custom connector button at the very bottom is what you want for LBM.

3. Fill in the Add custom connector dialog:

   • Name: LLM Brand Monitor
   • Remote MCP server URL: https://mcp.llmbrandmonitor.com/mcp
   • Leave Advanced settings collapsed — Claude discovers OAuth automatically via Dynamic Client Registration.
   • Click Add.

   

   Filled-in form. Anthropic surfaces the standard “Only use connectors from developers you trust” notice — same warning, slightly different wording from ChatGPT.

4. Claude opens a browser tab on llmbrandmonitor.com. Sign in if you’re not already logged in, then review the requested scopes on the consent screen and click Allow.

   

   The consent page is the same one ChatGPT users see — just with “Authorize Claude” at the top. Tick Remember my decision to skip this on subsequent reconnects.

5. You’re redirected back to claude.ai. Claude shows a Tool permissions panel where you can decide upfront when each tool is allowed to run.

   

   Claude’s defaults match LBM’s safety model: read-only tools (List Projects, Get Transcript, …) auto-run, write/delete tools (Run Scan, Archive Project, Delete Prompt, …) require explicit approval each time. Recommended: keep these defaults — Run Scan in particular spends credits, so always-approve protects you from accidental scans. You can revisit this any time at Settings → Connectors → LLM Brand Monitor.

6. In any conversation, open the + menu in the composer → Connectors → toggle LLM Brand Monitor on.

   

   The toggle is per-conversation — if you start a new chat, enable it again. Other connectors stay independent.

7. Send a prompt. Claude calls the LBM tools, asking for confirmation on any write/delete action you didn’t pre-approve.

   

   Example output for the prompt “Show me my LLM Brand Monitor projects and tell me which competitors AI models mention most often instead of my brand.” Claude returns project metadata plus a per-project competitor breakdown.

stdio (local) installation

All stdio clients run the same npm package — @serpstat/llm-brand-monitor-mcp — via npx. No global install required.

Prerequisites for every stdio client:

• Node.js 20 or newer — check with node --version. Install from https://nodejs.org if missing.
• An LBM_API_KEY from your LBM profile.

The first run downloads the package; subsequent runs are instant (npm caches it).

Claude Desktop (Windows & macOS)

Claude Desktop reads MCP servers from claude_desktop_config.json. The fastest way to open it is from inside the app itself.

1. Open Claude Desktop → Settings → scroll down to the Desktop app group → Developer. Click Edit Config in the Local MCP servers panel.

   

   The Edit Config button reveals claude_desktop_config.json in your default editor — no need to hunt for the file path manually. On older builds without this panel, open the file directly:

   OS	Path
   macOS	~/Library/Application Support/Claude/claude_desktop_config.json
   Windows	%APPDATA%\Claude\claude_desktop_config.json (typically C:\Users\<you>\AppData\Roaming\Claude\claude_desktop_config.json)

2. Add an lbm entry under mcpServers. If the file is empty, paste the whole snippet; if you already have other servers, just add lbm next to them and keep the JSON valid:

   {
     "mcpServers": {
       "lbm": {
         "command": "npx",
         "args": ["-y", "@serpstat/llm-brand-monitor-mcp"],
         "env": {
           "LBM_API_KEY": "lbm_your_key_here"
         }
       }
     }
   }

   

   Replace lbm_your_key_here with the key you copied from your LBM profile. Other top-level keys like preferences (if present) stay untouched — only mcpServers.lbm is what matters.

3. Save the file and fully quit Claude Desktop (right-click tray icon → Quit on Windows; ⌘Q on macOS), then relaunch it. A simple window-close is not enough — Claude only re-reads the config on a clean restart.

4. Verify the server is up: Settings → Developer → Local MCP servers. The lbm entry should show a green running badge.

   

   If you see a red status instead, click View Logs for the stderr output — most issues are missing LBM_API_KEY, JSON typos, or npx not on PATH (see the Windows note below).

5. Open a new conversation. Click the + in the composer → Connectors → toggle lbm on.

   

   Same per-conversation toggle as Claude Web — enable it again in any new chat. The 17 LBM tools become available immediately.

6. Ask Claude something — it will call the LBM tools and return live data.

   

   Example output for the prompt “Check my projects stats in LLM Brand monitor.” The “Loaded tools, used lbm integration…” line confirms the MCP server was actually called.

Windows note: if you use nvm-windows or a non-standard Node install, npx may not be on PATH for GUI apps. In that case, replace "command": "npx" with the absolute path, e.g. "C:\\Program Files\\nodejs\\npx.cmd".

Claude Code

Claude Code is Anthropic’s terminal CLI. It has a built-in mcp add command — the only thing to decide upfront is scope: where the server is registered and which Claude Code sessions can see it.

Pick a scope:

Tip: for the project scope, don’t hardcode the API key into .mcp.json — instead, reference an env var ("env": { "LBM_API_KEY": "${LBM_API_KEY}" }) and have each teammate export it in their shell. Otherwise the key ends up in git history.

Add globally (most common — recommended):

claude mcp add lbm-mcp -s user \
  -e LBM_API_KEY=lbm_your_key_here \
  -- npx -y @serpstat/llm-brand-monitor-mcp

Add only for the current project:

# run from inside the project directory
claude mcp add lbm-mcp \
  -e LBM_API_KEY=lbm_your_key_here \
  -- npx -y @serpstat/llm-brand-monitor-mcp

Add as a shared team config (commits .mcp.json to the repo):

# run from inside the project directory
claude mcp add lbm-mcp -s project \
  -e LBM_API_KEY='${LBM_API_KEY}' \
  -- npx -y @serpstat/llm-brand-monitor-mcp

Verify:

claude mcp list

You should see lbm-mcp in the output. Inside any Claude Code session, type /mcp to see live tool status:

The /mcp command groups servers by scope (Local MCPs, User MCPs, Project MCPs) and shows each one’s connection state. A green ✓ connected next to lbm-mcp means the npm package downloaded, the process started, and your LBM_API_KEY was accepted.

Press Enter on lbm-mcp to drill into the tool list:

All 17 LBM tools, tagged by Claude Code: read-only for safe lookups, destructive for tools that mutate state (Archive Project, Delete Prompt, and the credit-spending Run Scan). Claude Code asks for explicit approval before calling any destructive tool.

Use it: in any Claude Code session, just ask — Claude routes to LBM tools automatically.

Example output for the prompt “Show me my LLM Brand Monitor projects and tell me which competitors AI models mention most often instead of my brand. Eng please.” The header Called lbm-mcp 3 times (ctrl+o to expand) confirms the MCP server was invoked — press ctrl+o to inspect each call’s raw input/output.

Remove later:

claude mcp remove lbm-mcp           # removes from local scope (or whichever was used)
claude mcp remove lbm-mcp -s user   # removes the global registration
claude mcp remove lbm-mcp -s project

Codex CLI

OpenAI Codex CLI has a built-in codex mcp add command — same idea as Claude Code, just with slightly different flag names.

Add the server:

codex mcp add lbm \
  --env LBM_API_KEY=lbm_your_key_here \
  -- npx -y @serpstat/llm-brand-monitor-mcp

A few things to note:

• The -- separator is required before the command — everything after it is the launch command for the MCP process.
• --env takes a single KEY=VALUE; repeat the flag to pass multiple env vars.
• Codex writes the server into ~/.codex/config.toml as [mcp_servers.lbm] — you can hand-edit that file later if you prefer.

Verify:

codex mcp list

lbm should appear in the output. Inside any Codex session you can also run /mcp to inspect connection state and the available tools:

The /mcp panel shows every configured MCP server. For lbm you can verify the launch command (npx -y @serpstat/llm-brand-monitor-mcp), confirm the env contains LBM_API_KEY (masked as *****), and see all 17 tools enumerated as lbm_list_projects, lbm_run_scan, etc. Codex’s Auth: Unsupported line just means there’s no OAuth on stdio servers — the API key in Env is doing the auth.

Use it: ask Codex anything — it will route to LBM tools as needed.

Example output for the prompt “Show me my LLM Brand Monitor projects and tell me which competitors AI models mention most often instead of my brand. Eng please.” The Called lbm.list_projects(...) and Called lbm.list_competitors(...) lines are Codex’s tool-call trace — they’re your evidence that the MCP server actually ran.

Remove later:

codex mcp remove lbm

Manual config (alternative): if you’d rather edit the TOML directly — useful for committing a shared team config — add this block to ~/.codex/config.toml:

[mcp_servers.lbm]
command = "npx"
args = ["-y", "@serpstat/llm-brand-monitor-mcp"]
env = { LBM_API_KEY = "lbm_your_key_here" }

Save and restart any running codex session for the change to take effect.

Gemini CLI

Gemini CLI has a built-in gemini mcp add command — same idea as Claude Code and Codex, just with different positional syntax (the launch command and its args are passed inline, no -- separator).

Pick a scope:

Note: Gemini CLI’s default scope is project, the opposite of Claude Code (which defaults to local/cwd-only and gates global behind -s user). The end result for the default is similar — per-project — but the flag you need for global differs.

Add globally (most common — recommended):

gemini mcp add -s user \
  -e LBM_API_KEY=lbm_your_key_here \
  lbm npx -y @serpstat/llm-brand-monitor-mcp

The positional arguments are: <name> <commandOrUrl> [args...]. So lbm is the server name, npx is the command, and -y @serpstat/llm-brand-monitor-mcp are its arguments.

Add only for the current project:

# run from inside the project directory
gemini mcp add \
  -e LBM_API_KEY=lbm_your_key_here \
  lbm npx -y @serpstat/llm-brand-monitor-mcp

Verify:

gemini mcp list

Inside any Gemini session, run /mcp to see live connection state and the available tools:

The Configured MCP servers panel marks each server with a coloured dot — green means Ready, the process started cleanly and tools were discovered. For lbm you should see Ready (17 tools) with all 17 LBM tools listed underneath. Note Gemini’s tool-name convention: mcp_<server>_<tool>, so lbm_run_scan shows up as mcp_lbm_lbm_run_scan — Gemini uses these prefixed names internally, but you don’t need to type them, just talk to Gemini in natural language.

Use it: ask Gemini anything about your LBM account — it will pick the right tools automatically.

Example output for the prompt “Show me my LLM Brand Monitor projects and tell me which competitors AI models mention most often instead of my brand. Eng please.” The CSV-style block above the summary is raw lbm_list_competitors output (Gemini paginates at 20 rows by default — Showing 20 of 509. Use offset/page for next page.), and the prose below is Gemini’s synthesis with brand visibility per project.

Remove later:

gemini mcp remove lbm           # removes from project scope (default)
gemini mcp remove -s user lbm   # removes the global registration

Manual config (alternative): if you’d rather edit JSON directly, add this to ~/.gemini/settings.json (global) or .gemini/settings.json (per-project), then restart gemini:

{
  "mcpServers": {
    "lbm": {
      "command": "npx",
      "args": ["-y", "@serpstat/llm-brand-monitor-mcp"],
      "env": {
        "LBM_API_KEY": "lbm_your_key_here"
      }
    }
  }
}

OpenClaw

OpenClaw is a self-hosted automation runtime. It manages its own MCP config via the openclaw mcp subcommand — different from Claude Code / Codex / Gemini in that it doesn’t take separate --env and -- flags; instead, the entire server definition is passed as a single JSON object.

Add the server:

openclaw mcp set lbm '{
  "command": "npx",
  "args": ["-y", "@serpstat/llm-brand-monitor-mcp"],
  "env": { "LBM_API_KEY": "lbm_your_key_here" }
}'

The JSON shape is the same as the standard mcpServers.<name> block used by Claude Desktop and Gemini — openclaw mcp set just takes that object verbatim. Mind the shell quoting: single-quote the JSON to keep $, ", and newlines literal.

Verify:

openclaw mcp list           # one-line summary of every configured server
openclaw mcp show lbm       # full JSON for the lbm server
openclaw mcp show           # full MCP config (all servers)

openclaw mcp show reads ~/.openclaw/openclaw.json and prints the MCP block verbatim. The LBM_API_KEY value is masked (lbm_*****) — useful when sharing logs or screenshots. If you don’t see your lbm entry here, the openclaw mcp set command didn’t persist; double-check shell quoting on the JSON.

Use it: start an OpenClaw agent session and ask away — OpenClaw routes to LBM tools automatically.

Example output for the prompt “Show me my LLM Brand Monitor projects and tell me which competitors AI models mention most often instead of my brand. Eng please.” The status line at the bottom (agent main · session main · google/gemini-3.1-flash-lite-preview · tokens 162k/200k) shows OpenClaw running with Gemini as the underlying model — LBM tools work the same regardless of which LLM provider you wire OpenClaw to.

Remove later:

openclaw mcp unset lbm

Tip — bridge OpenClaw’s own channels into another MCP client: openclaw mcp serve exposes OpenClaw automation channels over MCP stdio. That’s orthogonal to LBM — you only need it if you want Claude/ChatGPT/etc. to drive OpenClaw itself. For LBM access, the set command above is all you need.

Verifying the connection

Once installed (HTTP or stdio), test it in any conversation with one of these prompts:

• Smoke test: What LLM Brand Monitor projects do I have? → the assistant should call lbm_list_projects.
• Credits: How many LBM credits do I have left? → calls lbm_get_usage.
• End-to-end: List my LBM projects, then show me the latest scan for the first one. → exercises lbm_list_projects + lbm_list_scans.

If the assistant says it doesn’t see the tools, see Troubleshooting.

Troubleshooting

For deeper debugging, run the server manually with the MCP Inspector:

LBM_API_KEY=lbm_your_key_here \
  npx @modelcontextprotocol/inspector \
  npx -y @serpstat/llm-brand-monitor-mcp

The Inspector opens a local web UI where you can list and call tools directly, which makes it easy to tell whether the problem is on the MCP side or in the client.

Need help? Open an issue at https://github.com/SerpstatGlobal/llm-brand-monitor-mcp/issues or contact support via your LBM profile.