The SEO and AI-visibility MCP server
Connect Claude, ChatGPT, Cursor or any MCP client to Sightkick, and your agent can see how AI answers your buyers’ questions, then research, write, score, schedule and publish against it. 31 tools, one URL, no API keys.
Your MCP endpoint
https://app.sightkick.so/mcpTwo minutes, whichever client you use
One endpoint for every client. The first call opens a sign-in and you pick your workspace on the consent screen — there is no key to paste.
Claude Code
Run this in your terminal — that's the whole setup.
claude mcp add --transport http sightkick https://app.sightkick.so/mcp
Claude- Open Settings → Connectors.
- Choose “Add custom connector” and paste the endpoint.
Same steps in the desktop app and on claude.ai.
ChatGPT- Open Settings → Connectors and turn on Developer mode.
- Choose “Add” and paste the endpoint.
Cursor- Open Settings → MCP.
- Choose “Add server”, pick streamable HTTP, and paste the endpoint.
VS Code- Run “MCP: Add Server” from the command palette.
- Pick HTTP and paste the endpoint.
OpenClaw- Add a new MCP server and paste the endpoint.
- Any other MCP client
- Add a custom (or remote) MCP server and paste the endpoint.
- Pick streamable HTTP if the client asks for a transport.
Anything that speaks MCP over streamable HTTP can connect — the endpoint is the same one every client above uses.
What your agent can do
Sightkick's own autopilot and your agent drive the same 31 tools. Neither needs the other: the product runs without an agent connected, and your agent can run it without touching the app.
Read what AI actually says about you
Every tracked buyer prompt, run daily against ChatGPT, Gemini, Google AI Overviews and AI Mode, with the stored answers behind it: whether you were mentioned, cited or recommended, which competitors won instead, and which pages the answer was built from.
visibility_summary, visibility_prompts, visibility_answer, visibility_sources, visibility_gaps
Read the search numbers, not a proxy for them
Real Google Search Console performance for the connected property — clicks, impressions, CTR and position over any range — plus the keyword set Sightkick researched for the site and the off-page coverage ledger.
metrics_gsc, keywords_list, outreach_list
Write it yourself, or hand it over
Your model can draft and edit articles directly in clean semantic HTML and grade every draft against the same five-pillar score the autopilot is held to — grounding, originality, AEO, intent, readability — iterating until it passes. Or delegate the whole staged pipeline and poll for the result.
articles_create_draft, articles_update, articles_score, articles_generate
Then actually ship it
Queue drafts, move the calendar, change the writing dial, take on or dismiss the to-do list, and publish to the connected CMS. Publishing takes an explicit confirmation unless the workspace is already running on full auto.
articles_queue, calendar_reschedule, autopilot_set_mode, actions_complete, publish_article
The full catalog, written for machines to read, is at /llm-info.
Write access you can hand over
Letting an agent publish to your site is only reasonable if the boundaries are real. These are enforced in the server, not documented as intentions.
- One grant, one website
- OAuth 2.1 with PKCE and dynamic client registration. The consent screen carries a workspace selector, and the grant is bound to the workspace picked there — an agent connected to one site cannot read or touch another.
- Every write is attributed
- Each call an agent makes lands in the workspace's activity feed against the client that made it. The person paying for the account can always see what their agent did, and when.
- Limits that survive a runaway loop
- Generation is capped at one in flight and five manual runs a day, scoring at 25 a day, writes at 20 an hour per workspace. Publishing needs an explicit confirmation that is meant to represent real approval, not an automatic retry.
- Included, never a separate tier
- Agent access ships with every paid plan. There is no developer add-on and no higher tier to unlock it — the meter is the work itself, not the way you ask for it.
Questions about the MCP server
What is the Sightkick MCP server?
A remote MCP server that gives any MCP client read and write access to one Sightkick workspace: AI-visibility tracking, Google Search Console data, keyword research, article writing and scoring, the content calendar, and publishing. It runs over streamable HTTP at a single URL — there is nothing to install or self-host.
Do I need an API key?
No. Authentication is OAuth 2.1 with dynamic client registration, so the first call opens a sign-in and you pick your workspace on the consent screen. Nothing to paste, nothing to rotate, and revoking access is a click in the app.
Which clients does it work with?
Anything that speaks MCP over streamable HTTP. Claude Code, Claude desktop and claude.ai, ChatGPT, Cursor, VS Code and OpenClaw are documented above; other clients use the same endpoint with their own add-a-server flow.
Can my agent write and publish, or only read?
Both. Your model can create and edit article content directly, score it against Sightkick's five-pillar judge, schedule it, and publish it to the connected CMS. Publishing is confirmation-gated unless the workspace already runs on full auto, and every write is attributed in the activity feed.
Does my agent's writing cost me article credits?
No. When your model writes the draft, it runs on your own subscription and Sightkick just provides the rails, the scoring and the publishing pipeline. Delegating to Sightkick's own generation pipeline is what draws on the workspace's allowance.
Is this a separate plan?
No. Agent access is included in every paid Sightkick plan. The product is also complete without it — the autopilot runs on its own, and connecting an agent is a way to drive or supervise it, not a requirement.
Give your agent something worth reading
Enter your website and Sightkick builds the picture — tracked prompts, keywords, competitors — before you connect anything.