MCP server · remote

Query your agent
analytics from Claude

The Serge MCP server connects your LLM client — Claude Desktop, Claude on the web, or Cursor — to your Serge agent-analytics workspace. Ask in plain language which AI agents visited your store, how much of that traffic was real buy-intent, and where agents failed to check out. The server reads your data; it does not change anything.

Connection guideGitHubScanner API docs
Jump
Connect in minutes

Mint an API key or add the connector, point your client at mcp.serge.ai, and ask your first question.

Jump
Read-only, scoped

Tools are scoped to your workspace and read your analytics data. Nothing is written or changed.

Jump
Eight analytics tools

Your client gets tools to look up scans, traffic, buy-intent split, verification, and failing sessions.

Connect your client

Two ways to authenticate. Pick one — you do not need both.

Option A — API key

Mint an API key in Settings → API keys. It starts with sk_serge_. Grant it the scans:read and traffic:read scopes.

Add this to your client config. The Claude Desktop config path is below; Cursor uses the same shape under its MCP settings.

macOS ~/Library/Application Support/Claude/claude_desktop_config.json
Windows %APPDATA%\Claude\claude_desktop_config.json

Restart your client. The Serge tools appear in the tool picker.

Option B — OAuth connector

In Claude, add https://mcp.serge.ai as a custom connector and sign in with your Serge account. Claude walks the OAuth flow and binds the connection to your workspace — no API key to paste or rotate.

Use this when you would rather log in than manage a key, or when you are connecting from Claude on the web.

Tools your client receives

Eight read-only tools, all scoped to your workspace. Your client picks the right one from your question.

whoami

Confirm which workspace and scopes the connection has. Call this first when a tool can't see your data.

list_sites

List the sites registered under your workspace, with their domains and IDs. Start here when you're unsure which domain a tool expects.

get_scan

Look up the latest scan for a domain — overall score, verdict, and a link to the human report.

get_traffic_overview

Summarize agent traffic on a site over a window: sessions, platforms (ChatGPT, Claude, Perplexity, Gemini), outcomes, and top entry pages.

get_purpose_split

Split agent sessions into buy-intent, informational, and crawler traffic, so you can tell real customer demand from bots.

get_verification_breakdown

Show, per platform, how many sessions cryptographically proved their identity versus only self-declared it.

find_failing_sessions

List sessions that abandoned a task or bounced, with the failure reason and entry/exit pages.

get_session_journey

Drill into one session — its page-by-page path, time per page, interactions, and outcome.

Scopes

An API key carries scopes that decide which tools it can call. Grant the least the question needs.

scans:readRead scan results for your domains. Required by get_scan.
traffic:readRead agent-traffic analytics. Required by the traffic, purpose, verification, and session tools.

Replay scopes exist internally but aren't grantable from the dashboard yet — coming soon.

Example prompts

Paste these into your connected client. It maps each to the right tool and fills in the rest.

How many AI agents failed to check out on yourstore.com this week?

What share of agent traffic on yourstore.com was real buy-intent versus crawlers in the last 7 days?

Is laptop-x-15 findable for agents on yourstore.com? Show me the latest scan.

Which platforms cryptographically verified their identity on yourstore.com, and which only declared it?

Troubleshooting

A tool can't see your data

Ask your client to call whoami. It returns the workspace and scopes on the connection — if the scopes are missing, mint a key with scans:read and traffic:read.

site_not_found

The domain isn't registered under your workspace, or you typed it differently. Ask your client to call list_sites and use a domain it returns.

How it fits with Serge

The MCP server is a read surface over the same agent-analytics data you see in the dashboard. It's the fastest way to ask a one-off question without clicking through filters.

When you need to watch a real agent attempt a purchase on your live store — and see exactly where it fails — that's the Agent Journey Test in the hosted product. The MCP server tells you what happened; the journey test shows you why.

Next

→ Open the step-by-step connection guide→ Open the Scanner API docs→ Run the public scanner