Docs

Connect an agent to your product feed

Connect SKULayer to Claude or another MCP client to work with the feeds in your workspace. Read audits, products, rules and Merchant Center issues. With explicit write access, create or change rules and start a run. Every rule returns an impact preview, and every action remains visible in the app.

What can a connected agent do?

Read tools work with every connection. A read-only connection can inspect the feeds and audit data in its authorised workspace and preview rules without changing data.

list_feeds

Every feed in the workspace with schedule, channels, item count, and the latest run summary.

get_feed

Full detail for one feed, including its rule stack size.

list_rules

The rule stack in execution order, each with a plain-English summary and how the rule reads in the app.

get_audit_findings

Audit findings for the latest run, grouped by check, with why each one costs money and how to fix it.

get_field_inventory

Every field your source actually sends, what we read each one as, and how much of your catalogue carries each shopping attribute before and after your rules.

get_merchant_issues

Google's own Merchant Center verdict on a connected feed: issue cases with affected items, states (open, fix shipped, cleared), and Google's documentation links.

search_items

Search or page through delivered items (what channels receive), or the base items before rules.

preview_rule

Test an unsaved rule against real items: match counts, field changes, warnings, and audit deltas. Writes nothing.

search_taxonomy

Search Google's product taxonomy: find the right googleProductCategory for a product type, or check a value before it goes into a rule.

get_run_status

One run's progress and how it finished, with item counts and the audit issue count.

Write tools need write access: approve it at the sign-in screen, or use a token created with the read-and-write scope.

create_rule

Create a rule. The full impact report comes back with it, so you see exactly what changes before the next run.

map_categories

Propose a Google category mapping over the feed's product types. It saves nothing; the mapping lands through create_rule once a human approves it.

set_rule_enabled

Switch one rule on or off. Takes effect on the next run.

delete_rule

Remove one rule permanently. Takes effect on the next run.

trigger_run

Start a feed run. A write tool, so it needs a paid plan and a read-and-write connection; the SKU limit and per-feed run cap apply.

Writes are preview-first

Nothing changes your feed silently. When an agent creates a rule, SKULayer runs the same preview every rule goes through in the app and returns the report with the created rule: how many items match, which fields change, what gets excluded, and any warnings, including the dangerous kind, like a rule that would empty a required field. The agent shows you that report in the conversation, so you confirm the impact where you are, phone included.

Rules created this way apply on the next run, so your live feed only changes when a new version passes every check. They stay visible in the app too: every one is listed on the feed page, marked as agent-created, and can be disabled or deleted in one click.

How do I connect?

Instructions vary by client. Pick yours. Claude signs in with your account; the others use an API token you create in the app under Account, then API tokens (workspace owners only). Pick read-only or read-and-write, and copy it when shown, it appears exactly once.

Claude (web & mobile) Claude Code Cursor VS Code ChatGPT Other clients

In Claude, go to Settings, Connectors, and add a custom connector with this URL:

https://skulayer.com/mcp

Claude opens a SKULayer sign-in. Sign in with your normal account, choose what the agent may do (read, or read and write), and approve. No token to copy, nothing secret in the URL, and access follows your account: sign-ins expire and refresh on their own.

The SKULayer plugin is the quickest route: it connects the server and adds skills for auditing, fixing, and watching your feeds.

claude plugin marketplace add SKULayer/skulayer-agent-plugin
claude plugin install skulayer@skulayer

When Claude Code asks, sign in with your normal SKULayer account and choose what the agent may do. Prefer just the server, no skills? Add it with an API token:

claude mcp add --transport http skulayer https://skulayer.com/mcp \
  --header "Authorization: Bearer skl_your_token_here"

Add SKULayer to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{
  "mcpServers": {
    "skulayer": {
      "url": "https://skulayer.com/mcp",
      "headers": { "Authorization": "Bearer skl_your_token_here" }
    }
  }
}

With MCP support enabled, add SKULayer to .vscode/mcp.json:

{
  "servers": {
    "skulayer": {
      "type": "http",
      "url": "https://skulayer.com/mcp",
      "headers": { "Authorization": "Bearer skl_your_token_here" }
    }
  }
}

ChatGPT supports remote MCP connectors. Add a custom connector pointing at the SKULayer endpoint with your API token:

https://skulayer.com/mcp
Authorization: Bearer skl_your_token_here

The exact menu path moves around between ChatGPT versions, so follow the current Connectors screen in your ChatGPT settings.

Point any MCP client at the endpoint over streamable HTTP with a bearer token:

https://skulayer.com/mcp
Authorization: Bearer skl_your_token_here

Try these first

Connected and staring at a blank box? Paste one of these. The first three are read-only and safe on day one; the rest need write access and are preview-first, so nothing changes until you say so.

List my feeds and summarize the audit findings on the worst one.

What is Google's Merchant Center complaining about on my feed right now?

Why did my last feed run finish the way it did? Walk me through the run status.

Fix the missing brands on my feed. Draft the rule, show me the preview, and don't save it until I say so.

Map my product types to Google categories, show me the proposed mapping, and stop before saving.

Fix the top audit finding, show me the preview, then run the feed and tell me when the audit is clean.

How is access controlled?

Every tool call is filtered to one workspace, however you connected. Sign-in access is scoped at the approval screen: read-only connections cannot reach the write tools at all, and removing the connection from your MCP client ends its access. API tokens are stored as hashes; we cannot read them back, and neither can anyone else. Each carries a scope, and revoking one from the Account page stops it immediately. The write tools also need a paid plan: on the free plan the connection is read-only. Even with write access, nothing applies without a preview report, a rule that trips a danger-level warning is saved turned off until you enable it, and every change is visible and reversible in the app.

Advanced · legacy clients

A client that can send neither a header nor complete a sign-in can append ?token=skl_your_token_here to the URL. URLs can land in logs, so treat that as the last resort: use a dedicated token and revoke it from Account when you're done.