Track Your AI Visibility from Claude Code and Codex

Short answer

How do I track AI visibility from Claude Code or Codex?

Create a Howseen API key, add the Howseen MCP server with one command, then ask your agent which buyer questions name a competitor instead of you.

The Howseen MCP server lives at https://app.howseen.ai/api/mcp and speaks Streamable HTTP, so both agents connect to it like any remote server. It gives your agent 26 tools over your Howseen data: the same visibility %, competitors and cited sources as your dashboard, for your own brands only. Read tools work with any key; action tools only run after a preview and your confirmation.

"MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems."

Why the terminal

Why developers and agencies want this data next to their code

The person who can fix a visibility gap (a page, a comparison table, a docs section) usually works in the repo. Putting the gap in the same terminal removes the dashboard round trip.

• From gap to pull request. Ask which question you lose, then ask the same agent to write or update the page that answers it, in your own codebase.
• Agencies, one key for every client. The list_brands tool returns every brand in your Howseen account, so you can review all clients in one session. See GEO for agencies.
• Real data, not guesses. Your agent reads the answers Howseen actually collected from ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode, instead of guessing what AI says about you.

If the idea of AI visibility is new to you, read what GEO (generative engine optimization) is and how AI chooses which brands to recommend first.

Real example

What it looks like on our own brand

On October 6, 2026, we asked Claude, connected through the Howseen MCP server, about Howseen itself, using our scan of October 5. The answer was not flattering, which is the point.

The useful part is the last number. The page AI cited most often in the answers we lost was a third party list article, so the first fix is to get listed there, not to rewrite our homepage. Different engines trust different sources: our study of 5,331 AI answers found YouTube cited in 45% of Google AI Overviews and in 0 of 1,164 ChatGPT answers. For how tools compare on this kind of data, see AI visibility tools.

Setup 1

Connect Claude Code

One claude mcp add command registers the Howseen server for your user account, so it is available in every project. Then type /mcp inside Claude Code to check it is connected.

1. Create a key. In Howseen, open Settings > "API & AI assistants" (app.howseen.ai/settings?tab=api). The key starts with hs_live_ and is shown once, so copy it right away.
2. Run this in your terminal, with your key in place of hs_live_YOUR_KEY:

claude mcp add --scope user --transport http howseen https://app.howseen.ai/api/mcp \
  --header "Authorization: Bearer hs_live_YOUR_KEY"

3. Check it. Start Claude Code and type /mcp. The howseen server should show as connected. We ran this exact command on October 6, 2026 and got "Connected".

Why --scope user: Anthropic's documentation says this scope stores the server in ~/.claude.json and makes it available across all your projects while keeping it private to your user account. The project scope, by contrast, is meant to be shared through version control, which is the wrong place for a key. Source: Connect Claude Code to tools via MCP.

Setup 2

Connect Codex

Codex reads the bearer token from an environment variable, so your key stays out of the Codex config file. Export the key, then run codex mcp add once.

export HOWSEEN_API_KEY=hs_live_YOUR_KEY
codex mcp add howseen --url https://app.howseen.ai/api/mcp --bearer-token-env-var HOWSEEN_API_KEY

Add the export line to your shell profile (for example ~/.zshrc or ~/.bashrc) so the key is still there in your next terminal session. To check the setup, run codex mcp list, or type /mcp inside Codex.

Prefer editing the file directly? The same server in ~/.codex/config.toml:

[mcp_servers.howseen]
url = "https://app.howseen.ai/api/mcp"
bearer_token_env_var = "HOWSEEN_API_KEY"

The --url and --bearer-token-env-var options and the url and bearer_token_env_var keys come from OpenAI's Codex documentation: Model Context Protocol (MCP) for Codex.

What to ask

7 prompts that work

You do not need to know the tool names. Ask in plain English and your agent picks the right Howseen tool on its own. Here are seven that work well.

Agencies can also use the client_report prompt (visibility, competitors and a share link) for each client brand. The full list of 26 tools is on the Howseen MCP page.

Permissions

Read only by default, actions only after you confirm

A standard key only reads. To let your agent change something, create a key with actions enabled; even then, every change comes as a preview and happens only after you confirm it.

Your plan limits apply exactly as in the app, and every applied change is listed in Settings under “Actions done by your AI”.

Security

Keep your key out of your repo

Treat a Howseen key like a password. Keep it in an environment variable or your user config, never in a committed file, and revoke it in Settings the moment it leaks.

• Never commit it. Do not paste the key into a project .mcp.json, a README or a script that goes to Git. With Claude Code, use --scope user; with Codex, use --bearer-token-env-var.
• Mind your dotfiles. If you sync ~/.zshrc or ~/.claude.json to a public repo, move the key to a file that is not synced.
• Avoid the URL form here. The ?key= option exists for clients that cannot send headers. Claude Code and Codex both can, so use the header or the env var.
• Revoke in one click. Each key can be revoked at any time in Settings > "API & AI assistants". Howseen keeps only a hash of your key, never the key itself.
• Rate limits. 120 requests per minute and 5,000 per day, per key.

Automation

Use it in CI or scripts

For a nightly job, a CI check or a client dashboard, skip the agent and call the read-only REST API directly, with the same key and the same data.

The REST API returns JSON with date range and engine filters and pagination. Store the key as a CI secret, never in the workflow file. Endpoints and a curl example are in the REST API section of the MCP page.

Pricing

What it costs

The MCP server is included on paid plans and during the 3-day free trial (card required). Howseen costs €79 per site per month, see brand pricing.

Not ready for a trial? Get a free AI report first: no card, about 2 minutes.