AI Visibility MCP for Continue

Continue runs in VS Code and in JetBrains, against whatever model you configured. Point it at AgentGEO and it gains one more capability that travels across both: fetch the raw answers six AI engines give for a query - ChatGPT, Perplexity, Google AI Overviews, Google AI Mode, Copilot and Gemini - with the answer text and citations arriving as structured data rather than a pasted screenshot.

Connect AgentGEO to Continue

Continue is driven by a config file. Add an mcpServers entry named agentgeo - the standard stdio server block:

{
  "mcpServers": {
    "agentgeo": {
      "command": "npx",
      "args": ["-y", "agentgeo-mcp", "--key", "ag_live_..."]
    }
  }
}

Nothing to build: the server is a zero-dependency npm package speaking MCP over stdio on Node.js 18+, and npx -y fetches it on first run. A live ag_live_ key comes with a plan. If you'd rather verify the wiring before spending a credit, an ag_test_ key returns clearly labelled demo records at zero cost. Reload Continue and fetch_raw_answers joins the assistant's tools - the same entry, and the same tool, in VS Code and in JetBrains.

Each call returns one record per requested surface - the provider's answer, verbatim, with every source it cited as structured data:

{
  "surfaceKey": "perplexity",
  "status": "delivered",
  "answerText": "...",
  "sources": [
    { "title": "...", "url": "https://...", "position": 1 }
  ],
  "fetchedAt": "2026-07-06T09:12:04Z",
  "latencyMs": 8412,
  "providerRecordId": "..."
}

What to ask once it's connected

No parameters to memorize - describe the outcome. Name a market and the assistant passes country through (it defaults to US). language defaults to en but only reaches Google AI Overviews - the chatbot surfaces take a country and no language field.

• "Fetch what chatgpt and perplexity answer for our main category query and tell me whether any cited domain is ours." The baseline check, in whichever IDE you happen to have open.
• "Run our homepage query across all six surfaces and write the citation lists into geo-notes.md." The evidence gets committed alongside the pages it judges.
• "Compare Gemini's cited sources with the files under content/ and tell me which page is closest to earning a citation."
• "Is our docs site in google_ai_overview's sources for our how-to query? If not, draft an opening paragraph that answers it in the first two sentences."
• "Diff copilot against chatgpt for our comparison query and list the competitors both of them name." Useful when the backend team lives in JetBrains and the frontend team lives in VS Code - same ask, same records.
• After the fix ships, re-fetch and diff answerText and sources[] against the record you saved last week. Check, fix, verify, no tab switching.

What the MCP server gives you

Thin on purpose. An open-source assistant shouldn't have to adopt a closed analytics product to answer a factual question about AI answers.

• One narrow tool - fetch_raw_answers. A single contract to learn, and one small addition to the assistant's context rather than a suite of tools.
• Six engines behind one contract - ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode and Copilot return an identical record shape, so widening a check is one more key in surfaces.
• Structured sources[] - title, URL and position for every citation, ready to diff, count or store. No links scraped out of prose.
• Raw records only - no rankings, sentiment or visibility scores baked in. The conclusions come from your assistant, running your model, on your repo.
• Managed-scraper engine, maintained for you - under the hood it posts to https://api.agentgeo.org/v1/fetches; nobody on your team babysits a headless browser or rotates proxies.
• Usage billing with a spend cap - one credit per delivered record, failed records cost zero, and never per-seat - so adding a JetBrains user costs nothing.

What it doesn't do

AgentGEO is a GEO data layer, not a platform. If you picked Continue to avoid handing your assistant's judgment to a vendor, you'll appreciate that this doesn't hand over the analysis either.

• No dashboards, no scores, no recommendations engine. Continue does the ranking and the diffing, with a model you chose and a repo it can read.
• Not instant. Live surfaces are slow: requests wait up to 180 seconds, and an AI Overview SERP round-trip alone runs 40-90s. A call across several surfaces waits on the slowest one.
• Some fetches finish async. A slow chatbot scrape can exceed the sync budget and return failed with a providerFields.snapshot_id; retry with that id and the same single surface to redeem the finished answer without paying for a re-scrape (not valid for google_ai_overview).

Prefer the REST API?

The MCP server wraps a single endpoint - POST https://api.agentgeo.org/v1/fetches - so anything you can ask Continue for, a script can do unattended. See the Python and curl pages; the contract is identical, which means a check born as a question in the assistant can move into CI without being rewritten.

Who builds on this

Teams that deliberately avoided a locked-in assistant - open source, bring your own model, and half the engineers in JetBrains while the other half live in VS Code. A GEO data layer that only worked in one editor, or only with one model, would recreate exactly the lock-in you rejected. This one is a config entry and a JSON contract, so it survives your next model migration and your next IDE argument. See where you stand first with the free audit, or see how the approaches compare.