IntegrationModel Context ProtocolOAuth 2.1

CitePatch MCP Server

Connect Cursor, Claude, Codex, VS Code, ChatGPT, or any MCP-compatible agent to your CitePatch workspace. Read visibility data, evidence, Brand Truth, technical health, shopping, and Search Console. It acts only within the permissions you grant. One-click OAuth, no API keys.

Add to your agent How it works

Endpointhttps://mcp.citepatch.com/mcp

Model Context Protocol

What is the CitePatch MCP server?

The Model Context Protocol (MCP) is an open standard that lets AI assistants talk to tools and data through one consistent interface. CitePatch exposes its visibility measurement, evidence, Brand Truth, technical, shopping, and Search Console data as an MCP server, so your agent can read the real picture without copy-paste. Reading is the default: a tool that changes something works only with a scope you tick.

It runs on Streamable HTTP at a single endpoint and authenticates with OAuth 2.1 — your client opens CitePatch consent, you pick the organization, and you approve exactly which scopes to grant. No API keys to create or paste. Nothing goes live from MCP: a page change opens as a pull request or a CMS draft that someone on your team reviews and merges.

Works with any MCP client — Cursor, Claude Code, VS Code, ChatGPT, Codex.
One-click OAuth — reads come ticked; every scope that changes something starts unticked.
/mcp/readonly serves the read tools only, for setups that must never write.
Evidence labeled honestly; external model text fenced as untrusted.

Your agents

Cursor
Claude Code
Codex
VS Code

OAuth 2.1 · PKCE

MCPmcp.citepatch.com

CitePatch

Visibility & prompts
Evidence & Brand Truth
Technical & citability
AI Shopping
Search Console

What your agent can do

You approve scopes at sign-in. Reads come ticked; the scopes that change something start unticked and work only if you tick them. Anything outside what you grant is rejected.

Workspace & projects

projects:read

See your workspace the way CitePatch sees it, then pick the right project.

List accessible projects and granted scopes
Pull a project's status briefing — domain, onboarding, last measurement

Visibility & prompts

measurement:read

Check AI visibility metrics and the prompts driving them.

Check AI visibility metrics for a project window
List tracked prompts and clusters, with gaps linked to each one
Read the AI answers observed for a prompt, sentiment included

Evidence & Brand Truth

evidence:read

Pull the real evidence behind a visibility gap and your approved facts.

Pull the full evidence chain behind a gap
Review approved Brand Truth facts for a project

Technical health & citability

measurement:read

Audit crawl health and citability from the latest finished crawl — never a live re-check.

Audit the Technical report — robots.txt, sitemap, JSON-LD, and more
Spot near-duplicate pages and the citations that repeat across them
Check citability — corpus coverage and page-level verdicts

Shopping & Search Console

measurement:read

Track AI Shopping performance and real Search Console data.

Track AI Shopping offerings, merchants, and findings
Look up one ranked AI Shopping product's detail
Pull Search Console rows — clicks, impressions, position

Changes, only with the scopes you tick

These start unticked on the consent screen, and /mcp/readonly does not offer them at all.

Tracked prompts

prompts:write

Change what CitePatch measures — only if you tick this scope.

Draft prompt suggestions for a project or one topic
Track a new prompt, or edit one's text and market
Pause a prompt, or stop tracking it

Page changes

patch:read · patch:generate · patch:apply

Draft the fix for a gap and open it for review. Nothing is ever published.

List page changes and follow one until it is ready
Generate a change for a gap — spends credits, touches nothing on your site
Open a ready change as a pull request or a CMS draft — never merged or published

Actions you took

actions:write

Tell CitePatch what you did outside it, so it can say whether it worked.

Record a forum reply, a listing or a profile you published (paid plans)
CitePatch then watches the prompts it affects and gives a verdict

Brand Truth facts

brandtruth:write

Keep the facts AI answers are checked against up to date.

Add a fact — suggested, or approved at once with the page that states it
Remove a fact — retracted or dismissed, its history kept

The built-in brief-me prompt walks the agent through a project briefing step by step.

Connect in 3 steps

Add the connector
Point your MCP client at https://mcp.citepatch.com/mcp — a CLI command in Claude Code, a JSON line in Cursor, or an HTTP entry in Codex. Configs below.
Sign in and approve
Your client opens CitePatch OAuth in the browser. Sign in, choose the organization, review the scopes — reads come ticked, changes do not — and Approve.
Start working
The client stores a short-lived token with the org bound in. Ask your agent about visibility, evidence, Brand Truth, technical health, shopping, or Search Console data — no headers to configure.

Wire it into your agent

Point any MCP client at https://mcp.citepatch.com/mcp (Streamable HTTP). The first call triggers CitePatch OAuth — pick an org, grant scopes, done. No API keys, no custom headers. https://mcp.citepatch.com/mcp/readonly is the same server without the tools that change something, for setups that must only read.

Install in Cursor Install in VS Code

Drop citepatch into your global MCP list, or commit a project file so the whole team gets the same connector.

In Cursor, open Settings → MCP and choose Add new global MCP server — or edit ~/.cursor/mcp.json directly.
Insert the JSON below, save, then reload MCP (restart Cursor if citepatch stays offline).
Authenticate when Cursor prompts you — Login beside citepatch, or the first tool call.
Optional: commit .cursor/mcp.json in the repo so teammates inherit the same endpoint.

.cursor/mcp.json

{
  "mcpServers": {
    "citepatch": {
      "url": "https://mcp.citepatch.com/mcp"
    }
  }
}

Desktop builds that only speak stdio need a local bridge — mcp-remote forwards HTTP and runs OAuth for you.

Open claude_desktop_config.json and add the citepatch block below.
Quit and reopen Claude Desktop so it picks up the server.
Approve CitePatch in the browser when the first tool runs, then confirm tools show in the UI.

claude_desktop_config.json

{
  "mcpServers": {
    "citepatch": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.citepatch.com/mcp"]
    }
  }
}

Any stdio-only client can reuse this npx mcp-remote wrapper against the same URL.

Run the CLI snippet below from a terminal.
Inside Claude Code, open /mcp, pick citepatch, and choose Authenticate.
Finish consent in the browser, then check with claude mcp list.

terminal

claude mcp add --transport http citepatch https://mcp.citepatch.com/mcp

Works for user-level MCP and workspace configs — including Copilot Chat when MCP is enabled.

Command Palette (Cmd+Shift+P / Ctrl+Shift+P) → MCP: Open User Configuration.
Merge the JSON below into mcp.json, or place it under .vscode/mcp.json for the workspace.
MCP: List Servers → start citepatch → finish OAuth when VS Code asks.

mcp.json

{
  "servers": {
    "citepatch": {
      "type": "http",
      "url": "https://mcp.citepatch.com/mcp"
    }
  }
}

Schema quirk: top-level key is servers, and each entry needs type: "http" (not the mcpServers shape Cursor uses).

Add CitePatch as a connector in ChatGPT settings — URL plus OAuth, nothing to paste as a secret.

Go to chatgpt.com/#settings/Connectors.
Add a connector and paste the CitePatch URL below.
Choose OAuth, sign in to CitePatch, and grant the scopes you need.

server URL

https://mcp.citepatch.com/mcp

Declare an HTTP MCP server in your Codex config, then reload so discovery can run.

Append the TOML block below to your Codex MCP config.
Reload MCP (or restart Codex) and complete OAuth if the client opens a browser.
Sanity-check with a read tool such as citepatch_get_workspace.

config.toml

[mcp_servers.citepatch]
url = "https://mcp.citepatch.com/mcp"

OAuth over MCP is still uneven across Codex builds — if discovery stalls, Cursor or Claude Code are the safer path.

Try it

Once connected, talk to your agent in natural language. It picks the right tools.

Brief

“Use citepatch_get_workspace then brief me on our top visibility metrics for the main project.”

Diagnose

“List our tracked prompts and show the evidence chain for the highest-impact gap.”

Audit

“Pull our Technical report and flag anything blocking AI crawlers.”

“Show our Search Console performance for the last 30 days, sorted by impressions.”

Built-in guardrails

Protections

OAuth 2.1 with PKCE — short-lived tokens, revocable instantly in Settings → MCP.
Org isolation — organization is signed into the token; tenant args are rejected.
Changes only with the scopes you tick, and nothing goes live: a page change opens as a pull request or a CMS draft for your team to review.
Honesty contract — evidence labels, metric cells, untrusted fences on external text.

Privacy & data

MCP calls send your AI client exactly what each tool returns — no raw embeddings dump, no internal metadata. Every call is scoped to your own token and leaves an audit trail. We do not use your workspace data to train third-party models.

Connect CitePatch to your AI agent

Add the endpoint, approve the scopes you want in one click, and give your agent the real evidence behind your AI visibility.

Add to your agent Create a free account

CitePatch MCP Server

Workspace & projects

Visibility & prompts

Evidence & Brand Truth

Technical health & citability

Shopping & Search Console

Tracked prompts

Page changes

Actions you took

Brand Truth facts

Connect in 3 steps

Add the connector

Sign in and approve

Start working

Built-in guardrails

Protections

Privacy & data

Connect CitePatch to your AI agent