We've released an MCP (Model Context Protocol) server for Feedoptimise. If you use Claude or ChatGPT, you can now connect your Feedoptimise account and manage your product feeds by chatting instead of clicking around the UI.
What MCP is
MCP is an open standard that lets AI assistants talk to external systems through a consistent interface. The assistant adds a connector, authenticates, and from then on it can call tools on that server as part of a conversation. Both Claude and ChatGPT support MCP connectors.
We've been running an AI Feed Agent on Feedoptimise for months, and it's changed how teams investigate issues, audit feeds, and create new rules and strategies. This release exposes the same capabilities to customers through Claude, ChatGPT, or any other MCP-compatible client.
What you can ask it to do
Some examples of things that work today:
- "Why is SKU 48291 disapproved in Google Merchant Center?"
- "Run a full Google Shopping audit on my UK feed."
- "My Shopify source imported 780 items today instead of the usual 1,200 - what happened?"
- "Force a re-import of the Magento source."
- "Who changed the mapping on the Google feed this week?"
- "Block SKU 48291 item from the Google and Meta feeds"
- "How do I create a custom label based on Google Ads conversion data?"
- “Explain to me the custom labels rules in the TikTok feed”
The AI picks the tools, chains the calls, asks for confirmation before anything destructive, and summarises what it found.
What's available
You don't need to memorise tool names. The AI picks the right ones based on what you ask. Here's what's exposed:
Sources (imports) List and inspect all import sources (Shopify, Magento, WooCommerce, BigCommerce, PrestaShop, URL, FTP, SFTP, API). View mapping, schedule, and settings. Force a re-import. Change import schedules. View analyser results and blocked-item counts. Audit change history.
Feeds (exports) List and inspect all export feeds (Google, Meta, Amazon, TikTok, Bing, custom channels). View mapping, format, connectors, and delivery settings. Force a sync and push to delivery targets. View analyser results and processing statistics. Run compliance audits against channel specifications. Audit change history.
Items (products) Search the catalog by title, SKU, MPN, GTIN, or ID. View raw source data. View how an item renders in a specific feed. Fetch landing-page metadata (schema.org, pixel data, Open Graph). Block or unblock items at source level or feed level.
Google Merchant Center List connected GMC accounts. Pull account-level, feed-level, and item-level diagnoses. Cross-reference feed data against landing page data to find mismatches.
Support and knowledge Semantic search across the full Feedoptimise help center. Guide on how to set up rules and more. Pricing and billing questions.
Connecting Feedoptimise
The rest of this page is the connection documentation. The steps below use Claude as an example, but the same server URL and auth flow works with ChatGPT and any other MCP-compatible client.
Prerequisites
- A Feedoptimise account on any plan
- Access to Claude.ai (Pro, Team, or Enterprise) or ChatGPT (Plus, Team, or Enterprise) - any plan that supports MCP connectors
Step 1 - Add the connector
In Claude.ai: Go to Settings → Connectors, click Add custom connector, and enter:
- Name: Feedoptimise
- Server URL: https://api.feedoptimise.com/v1/mcp/handle
- Transport: HTTP
Click Connect.
In ChatGPT: Go to Settings → Apps → Advanced settings → Enable Developer mode → Create app - add a new MCP server, and use the same server URL above.
Step 2 - Authenticate
You'll be redirected to Feedoptimise to authorise the connection. Log in and approve access. The flow is OAuth2. A single connection covers all Feedoptimise accounts your user has access to - there's no need to add separate connectors per account.
Step 3 - Use it
Open a new chat. At the start of each session, the agent will list the Feedoptimise accounts available to you and ask which one to work with. Pick an account, and everything from that point operates within a single account. You can switch to a different account mid-conversation by asking.
On the first call the server also returns an instructions payload that gives the AI the platform context it needs, so there's nothing you have to configure beyond selecting an account.
Security
- Auth: OAuth2 with scoped access tokens. A single connection gives access to all accounts your user belongs to; account selection happens at session start.
- Permissions: The MCP server respects your Feedoptimise role per account. If you're read-only on an account, you're read-only through MCP on that account.
- Revocation: Revoke anytime from Feedoptimise (Account → Credentials Hub → Type: OAuth2) or from the connector settings in Claude/ChatGPT.
- Data scope: You only see accounts your Feedoptimise user has access to. Each session operates against one selected account at a time.
- Audit trail: Every change made via MCP is logged in Feedoptimise's standard change history, attributed to the user who authorised the connection with an MCP tag.
Destructive actions
The agent will always describe what it's about to do and wait for explicit confirmation before:
- Blocking or unblocking items
- Forcing imports or feed syncs
- Changing schedules
- Any other destructive actions
Troubleshooting
"Connection failed" when adding the connector. Verify the server URL is exactly https://api.feedoptimise.com/v1/mcp/handle.
"Authorisation expired." Tokens refresh automatically, but if you see this, remove the connector and add it again. It's usually caused by a password change or revoked session on the Feedoptimise side.
Can't find a feed, source, or item. Check you're working on the right account, ask "which account am I connected to?" to verify, or ask to switch accounts.
Permission errors on actions. Some actions require admin permissions in Feedoptimise. Check your role under Account → Team.
Support
- Help Center: https://app.feedoptimise.com/help/center/
- Support tickets: open from the Feedoptimise app
- Contact form: https://www.feedoptimise.com/contact
Roadmap
We'll keep expanding the server based on what customers actually use. On the list:
- Bulk item actions (block/unblock by analyser label)
- Conversational mapping edits
- Deep analysis of performance reports (Google Ads, Analytics)
If there's a workflow you'd like supported, contact us.