Edit your product feeds with Claude.

The dashboard is built for designers and marketers, it's good at one feed at a time. The MCP server is built for language: you describe what you want, Claude makes it happen across feeds, with brand awareness, and with verification baked in.

• Brand-aware templates.Put your brand specs (colors, fonts, voice, do-not-use words) in your project's CLAUDE.md. Ask Claude to design a catalog image, and it generates HTML/CSS that matches first try, without you pasting tokens.
• Inline preview. Claude calls preview_render and sees the rasterised PNG/JPEG inline. It iterates visually, not blindly.
• Validation-driven rules. Claude calls get_validation, sees what Meta / Google flags, and writes the exact update_field_rules call to fix it. The infamous title-eats-description bug? One sentence to Claude.
• Bulk-action language."Add a brand prefix to every feed targeting Heureka", one prompt, three tool calls, done. Same task in the UI is one feed at a time.

Every tool is scoped to the connected account. You can't accidentally touch another account's feeds, whether you connect via OAuth or a bearer key.

Emberfeed connects through Claude's built-in connector UI, no config files and no tokens to paste. Claude signs you in to Emberfeed with OAuth and stores the connection for you.

1. Open the connector dialog. In Claude Desktop or on claude.ai: Settings → Connectors → Add custom connector (the + next to the connector list).
2. Point it at Emberfeed. Name it Emberfeed and paste the server URL:

   https://emberfeed.com/mcp

   Leave the OAuth Client ID / Secret fields blank, Emberfeed registers your client automatically.
3. Sign in. Click Add, keep Individual sign-inselected, and approve. You sign in to your Emberfeed account once, there's no API key to copy or store.
4. Use it.Ask "list my Emberfeed feeds". All 21 tools are now available (set them to Needs approval or Always allowin the connector's tool list).

Claude Code, or a manual bearer token

On the CLI, add the server and let Claude Code drive the same OAuth flow, then run /mcp and authenticate when prompted:

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

For headless or scripted use where no browser sign-in is possible, issue a key at /account/api-keys and pass it as a bearer header instead:

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

Two layers of context make this work. The Emberfeed layer is automatic: the server tells Claude to call get_brand_context_starter at the start of a session, so it already knows every Handlebars helper, every per-platform spec, and four starter template patterns before you say a word.

The your-brandlayer is the part only you have. Put it where Claude looks first: the feed's Settings → Brand context field. Claude reads it over MCP with get_feed and writes it with update_feed_brand_context, so your colors, fonts, voice, and image rules travel with the feed, no project file required. The same notes also work in a project's CLAUDE.md if you prefer:

## Emberfeed brand

- Primary color: #FF6600 (ember orange)
- Display font: Space Grotesk
- Body font: Inter
- Voice: confident but practical, no buzzwords
- Required on every catalog image: small logo top-left,
  price chip top-right, gradient overlay on bottom 40%
- Square (1:1) for Meta feed placement.

Then prompt Claude: "Design a new code template for my feed; the brand is in this project's CLAUDE.md. Render against the first product and show me." Claude calls create_code_template, then preview_render, and you see the actual rasterised image in chat. Iterate by talking. No CSS files in the editor.

Free tier: 10 MCP calls per minuteper API key. Paid: 60/min. Tools don't count against the daily render budget, only preview_render and the served XML consume the regular per-user budget defined by your plan. Rate-limit headers (x-ratelimit-remaining) come back on every response so Claude can self-pace.