New · MCP server + REST API
Howseen MCP: See Where AI Picks Your Competitor Over You
Ask Claude, ChatGPT or Cursor which buyer questions name a rival instead of your brand, then fix it from the same chat. We ran it on our own brand first: on October 5, 2026, 45 of our 50 tracked buyer questions named a competitor and not us.
Every change is shown to you first. Nothing happens until you confirm.
Claude & Claude Code
ChatGPT & Codex
Cursor
By Raphaël Aubry, founder of Howseen AI · Updated Overview
What is the Howseen MCP server?
The Howseen MCP server lets Claude, ChatGPT or Cursor read your Howseen AI visibility data and, when you allow it, act on it. Keys are read only by default; nothing changes until you create a key with actions enabled and confirm each change.
Howseen tracks how ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode answer your buyers' questions: whether they name your brand, which competitors they name instead, and which sources they cite. If you are new to this, start with what GEO (generative engine optimization) is and how AI chooses which brands to recommend.
Until now, that data lived in the dashboard. The MCP server brings it into the AI assistant you already use. It speaks the Model Context Protocol, the open standard that AI apps use to plug into outside data:
"Think of MCP like a USB-C port for AI applications."
Anthropic open-sourced the protocol in November 2024 (announcement), and OpenAI documents how its models connect to remote MCP servers (OpenAI MCP guide). The numbers your assistant gets from Howseen are the same as the ones in your dashboard, for your own brands only.
Use cases
See the gap. Then close it.
Every buyer question you lose is a recommendation that went to someone else. Your assistant shows where ChatGPT, Gemini, Perplexity and Google AI name your competitors instead of you, which sites those answers trust, and helps you win the question back.
| You ask | What happens |
|---|---|
| "Which buyer questions name my competitors and not me?" | It lists the exact questions where you lose, engine by engine |
| "Which sites should I get listed on first?" | It ranks the sites AI already cites in those answers, easiest win first |
| "What changed since my last scan?" | It tells you what moved and the likely reason |
| "Write the article that wins back my biggest gap, then publish it." | It drafts the article, shows it to you, and publishes only once you say yes |
New here? Start your 3-day free trial, the MCP server is included. Already on Howseen? Create your key in Settings > API & AI assistants.
Proof
The proof, run on ourselves
On October 5, 2026, across 50 real buyer questions, a competitor was named in 45 and Howseen was not. Our visibility was 4%.
We do not have a client case study to show yet, so we pointed Howseen at Howseen. The same answer did not stop at the damage: it returned a ranked list of the third party sites already cited in those lost answers, the ones to work on first.
For the wider pattern, our public study of 5,331 AI answers found YouTube cited in 45% of Google AI Overviews and in 0 of 1,164 ChatGPT answers. Different engines trust different sources, which is why you need to see each one.
Reference
The 26 tools
Read tools work with any key. Action tools need a key with actions enabled, and every change is previewed before you confirm it. Your assistant picks the right tool on its own, based on the question you ask.
| Tool | What it does |
|---|---|
| See the gap (any key) | |
get_changes | What moved between your last two scans, per AI engine |
get_prompt_gaps | Questions where a competitor is named and you are not |
get_easiest_wins | The sites AI cites in those answers, ranked, easiest first |
get_action_plan | Your next best actions |
get_visibility | Visibility %, share of voice, average position and trend, per engine |
get_competitors | Competitor ranking by visibility |
get_top_sources | Most cited domains and URLs, per engine |
search_responses | Real AI answer excerpts with their citations |
list_prompts | The buyer questions you track |
list_articles | Articles Howseen wrote for the brand |
get_article_performance | How your articles index on Google and Bing |
suggest_prompts | Real buyer questions you do not track yet |
get_usage | Your plan, quotas and what your key may do |
list_brands | The brands tracked in your account |
| Close the gap (key with actions enabled, previewed then confirmed) | |
add_prompts, archive_prompts, remove_prompts | Choose the questions you track |
add_competitor, remove_competitor | Choose who you compare against |
mark_action_done, dismiss_action | Keep your action plan current |
draft_article | A draft that answers one buyer question (never published by itself) |
run_scan | Refresh your data, same limit as the Run now button |
create_share_report | A client-ready report link with your agency branding |
| Publish (previewed then confirmed) | |
publish_article | Publish a draft to your connected CMS (WordPress, Shopify, Ghost, BigCommerce, webhook) or your Howseen hosted blog, now or on a date |
request_indexing | Ask Google and Bing to index the new page |
Ready-made prompts: weekly_review (what changed, your gaps, your easiest wins, what to do this week) and client_report (visibility, competitors and a share link).
Setup
Set it up in 2 minutes
Create a personal API key in Howseen, paste the ready-made setup for your AI app, then ask your first question. No code to write.
- Create a key. In the Howseen dashboard, open Settings > "API & AI assistants" (app.howseen.ai/settings?tab=api). Your key starts with
hs_live_and is shown once, so copy it right away. It is read only unless you enable actions. - Copy your app's setup. Settings has a copy-paste setup for Claude, Claude Code, ChatGPT, Codex, Cursor and REST, with your key already inside.
- Ask. Try "Which buyer questions name my competitors and not me?" and your assistant answers with your Howseen data.
| App | Connection |
|---|---|
| Claude | One custom connector with the server URL, then it works on claude.ai, in the Claude Desktop app and on mobile |
| Claude Code | One command in your terminal (claude mcp add), available in every project |
| ChatGPT | MCP connector, in developer mode |
| Codex | One command in your terminal (codex mcp add), key read from an environment variable |
| Cursor | mcp.json entry (Settings generates it for you) |
| Scripts | REST API with curl or any HTTP client |
The server lives at https://app.howseen.ai/api/mcp (Streamable HTTP). It authenticates with the header Authorization: Bearer <your key>. For clients that cannot send headers, the key goes in the URL as ?key= instead. Some AI apps can only send your key inside the server URL: treat that URL like a password, do not share it or show it on screen, and revoke the key in Settings if it leaks.
For developers and agencies
REST API for developers and agencies
The same data is available as a read-only JSON REST API, with date-range and engine filters and pagination, for your scripts, reports and client dashboards.
| Endpoint | Returns |
|---|---|
GET /api/v1/brands | Your brands |
GET /api/v1/brands/{id}/visibility | Visibility, share of voice, position, trend |
GET /api/v1/brands/{id}/prompts | Tracked buyer questions |
GET /api/v1/brands/{id}/competitors | Competitor ranking |
GET /api/v1/brands/{id}/sources | Most cited domains and URLs |
GET /api/v1/brands/{id}/responses | AI answer excerpts and citations |
GET /api/v1/brands/{id}/actions | Next best actions |
GET /api/v1/brands/{id}/articles | Articles |
curl "https://app.howseen.ai/api/v1/brands/BRAND_ID/visibility?from=2026-09-01&to=2026-09-30&engine=chatgpt" \
-H "Authorization: Bearer hs_live_YOUR_KEY"- Filters:
fromandto(YYYY-MM-DD),engine(chatgpt, gemini, perplexity, ai_overviews, ai_mode). - Pagination:
limitandoffset. - Rate limits: 120 requests per minute and 5,000 per day, per key.
Agencies can pull every client brand into their own reporting. See GEO for agencies and the agency plans.
Trust
Security and privacy
Read only by default. A key can only act if you create it with actions enabled, and every change needs your confirmation. Each key only sees your own brands, and Howseen never stores your key itself, only its sha256 hash.
- Read only by default. A key can only act if you create it with actions enabled, and every change needs your confirmation. The REST API stays read only.
- Every change is logged. Applied changes are listed in Settings under “Actions done by your AI”, and your plan limits apply exactly as in the app.
- Your data only. A key is scoped strictly to the brands of the account that created it, or to a single brand if you choose.
- Key shown once. Howseen keeps only a sha256 hash, so nobody can read your key back, including us.
- Revocable. Each key can be revoked at any time in Settings.
Questions, answered
What is an MCP server?+
An MCP server exposes data and tools through the Model Context Protocol, an open standard for connecting AI applications to external systems. The Howseen MCP server exposes 26 tools over your AI visibility data: read tools for your numbers, and action tools (track questions, draft and publish articles) that only run after you confirm.
Can my assistant change my account?+
Only if you allow it. API keys are read only by default. To let your assistant act, create a key with actions enabled. Even then, every change comes as a preview first (what will change and the effect on your quotas) and only happens after you confirm. Your plan limits apply exactly as in the app, every applied change is listed in Settings under “Actions done by your AI”, and you can revoke a key at any time.
Which apps are supported?+
Any assistant that supports remote MCP servers. Settings gives you a ready-to-copy setup for Claude (claude.ai, the Claude Desktop app and mobile, through one custom connector), Claude Code (one terminal command), ChatGPT (developer mode connectors), Codex (one terminal command) and Cursor (mcp.json). Developers can also read the same data from the REST API.
Is my data safe?+
Each key only sees the brands of the account that created it. Howseen stores only a sha256 hash of your key, shows the key once, and lets you revoke it at any time. Some apps can only send your key inside the server URL: treat that URL like a password.
Do I need to code?+
No. You create a key in Settings, copy the setup for your AI app, paste it, and ask questions in plain English. Claude Code and Codex take one command in your terminal. The REST API is for developers who want to script.
Are there limits?+
Yes, so nothing overloads your plan, and normal use stays well under them: 120 requests per minute and 5,000 per day per key. Scans follow the same Run now rule as the app, and your data refreshes automatically every 3 days.
Which plans include it?+
The MCP server and the REST API are available on all paid Howseen plans and during the 3-day free trial.
Can agencies use it for several brands?+
Yes. The list_brands tool returns every brand tracked in your Howseen account, so one key covers all your clients' brands, and create_share_report gives you a client-ready report link with your agency branding. You can also create a key limited to one brand for a client.
You cannot fix a gap you have not measured
Connect your assistant for 3 days, free, and ask it who gets named instead of you. Included on all paid plans, see brand pricing.
