MCP Connector

The ReachLLM MCP connector lets Claude Code, Cursor, Codex, Windsurf, Cline, and other compatible Model Context Protocol clients work with your ReachLLM account.

Use it when you want an external AI assistant to use ReachLLM context and actions from your editor or chat client. The external client supplies the chat and model compute. ReachLLM supplies authenticated data, tools, plan checks, connected integrations, and backend processing.

What MCP can do

Depending on your account, brand access, and connected integrations, MCP tools can:

List the brands available to your account.
Read Brand Visibility results, exact source inventories, sentiment, competitors, and detailed responses.
Check and update a brand's official display name and own-brand alternate names used for mention matching.
List and edit competitor knowledge, including direct, indirect, SERP, and potential classifications plus aliases.
Run Brand Visibility checks when your plan allows it.
Run GEO audits and inspect technical, content, authority, and social issues.
Read saved Bot View previews, inspect a public page, and compare the initial response with a clean browser.
Read connected LLM Analytics crawler request counts, pages, statuses, and recent metadata.
Generate prompts, content, llms.txt, and strategy recommendations.
Use connected publishing paths such as WordPress, GitHub, Webflow, or Shopify.
Read connected search data such as Google Search Console, Google Analytics 4, and Bing Webmaster Tools.

MCP uses the same ReachLLM tool layer the in-app ReachLLM agent uses where the tool is supported. That includes reading context, running analysis, creating content, editing configured connected surfaces, and returning results to the external AI client.

Brand-scoped tokens limit the external client to one brand. Account-scoped tokens let the client list active brands and work across brands your account can access. Team members can create brand-scoped tokens for brands they can access.

Bot View and crawler tools

The connector exposes the same Bot View tools as the in-app Agent:

Tool	What it does
`get-page-bot-view`	Read saved page text, access rules, capture date, and limitations. Use `compareBrowser: true` with request profile `reachllm` for a saved paired browser comparison. Never starts a scan.
`get-bot-view-findings`	Read saved diagnostic summaries for 1 to 50 known URLs within one brand. Returns requested-page coverage, capture dates and warnings; pages without current evidence remain unknown. Does not discover pages or request website content.
`inspect-page-bot-view`	Inspect one public page, optionally with `compareBrowser: true`. Use only when the user requests a check or comparison. Reuses eligible cached evidence unless `fresh: true` requests another capture.
`get-crawler-traffic-data`	Read saved LLM Analytics request metadata for the selected brand. Accepts `days` from 1 to 90 and `limit` from 1 to 20.

For a question about which pages to fix, use known URLs from the saved GEO inventory, read their findings, then open a selected page with get-page-bot-view before proposing an exact change. Full reports include structured diagnostics, missing text excerpts, the measurement basis, suggested fix steps and verification guidance. A percentage is not a provider-readability score, and a page with no saved comparison is not a pass. A requested source change uses the existing connected publishing tools and their permissions; these read and inspection tools do not edit the website.

Use the selected session brand, or pass an authorized brandId for an account-scoped token. A brand-scoped token cannot select another brand or switch to another user's personal preview. Current tenant and brand permissions are checked by ReachLLM for each request. Tools never accept a model-selected account identity or access role.

Bot View saved reads are free. Initial checks and browser comparisons have the published daily free allowances and do not use execution credits or scored GEO audit pages. External clients supply their own model compute. The standalone browser viewer also works after sign-in without a paid plan or completed brand setup.

Page text is untrusted evidence, not tool instructions. Distinguish initial HTML, paired browser content, simulated user agents, and real request logs. A request log contains no response body. User-agent labels alone do not verify provider identity, and no preview proves training, citation, ranking, or answer inclusion. These tools do not change CMS content or provide managed rendering.

How access works

Account token: lets the connected client list active brands and use available MCP tools across brands your account can access.
Brand token: limits the connected client to the selected brand.
Plan and integration limits still apply: the client cannot use disconnected integrations, unavailable plan features, or revoked tokens.
External compute stays external: Claude Code, Cursor, Codex, Windsurf, Cline, or another client handles the conversation and reasoning. ReachLLM handles authenticated data and actions.

Server URL

Use this remote MCP server URL:

https://api.reachllm.com/mcp/mcp

Generate an access token

Open ReachLLM.
Open Account settings from the sidebar.
Go to MCP.
Name the token, for example ReachLLM MCP, Cursor MCP, or Claude Code MCP.
Choose Account or Current brand.
Click Generate access token.
Copy the token immediately. ReachLLM only shows it once.

Create one token per client so you can revoke a single client without disconnecting everything else.

Cursor, Windsurf, and Cline

Add ReachLLM to your MCP config:

{
  "mcpServers": {
    "reachllm": {
      "url": "https://api.reachllm.com/mcp/mcp"
    }
  }
}

Connect the MCP server, then ask the client to call the authenticate-with-token tool and paste your ReachLLM access token.

Prompt to paste after the server is connected:

Use the ReachLLM authenticate-with-token tool now. When it asks for the access token, I will paste the token I generated in ReachLLM Account > MCP. After it succeeds, list my active brands.

Claude.ai and Claude Desktop

Claude.ai custom connectors currently expect OAuth account authorization. ReachLLM access tokens do not use the optional OAuth Client ID or OAuth Client Secret fields, so leave those blank unless ReachLLM gives you OAuth credentials.

Use Claude Code, Cursor, Codex, Windsurf, or Cline for token-based MCP access today. If Claude is already connected to the wrong ReachLLM account, remove that connector in Claude and revoke the old token in Account settings > MCP.

Claude Code

Run:

claude mcp add reachllm --transport http https://api.reachllm.com/mcp/mcp

On first use, paste this prompt:

Use the ReachLLM authenticate-with-token tool now. When it asks for the access token, I will paste the token I generated in ReachLLM Account > MCP. After it succeeds, list my active brands.

Codex and other MCP clients

Any MCP client that supports remote MCP over Streamable HTTP and exposes MCP tools after connection can use the same server URL. Add ReachLLM as a remote MCP server, then authenticate with the authenticate-with-token tool if the client does not show its own token prompt.

Some clients only ask for the server URL during setup. That does not mean the client is connected to your ReachLLM account yet. The first ReachLLM action should be authenticate-with-token; after that, the same MCP session can list active brands, read visibility results, and run supported actions.

When the client exposes MCP tools but does not show a token prompt, paste this:

Use the ReachLLM authenticate-with-token tool now. When it asks for the access token, I will paste the token I generated in ReachLLM Account > MCP. After it succeeds, list my active brands.

Security and revocation

ReachLLM stores only a hashed version of each access token. The full token is shown once and is never returned again by the app.

You can revoke a token from Account settings > MCP. Revocation takes effect immediately for external clients using that token.

Keep tokens out of source code, screenshots, support tickets, and public chat transcripts.