MCP

The Cevoid MCP server is Cevoid's hosted Model Context Protocol⁠ server. It gives AI clients such as Claude, Cursor and ChatGPT secure access to your Cevoid workspaces through a single endpoint.

Through it, a client can query UGC, Loyalty and Profiles analytics, manage reports and dashboards, and edit, preview and publish the BrandKit that reviews and emails use: styling, color schemes, fonts and custom SVG icons.

Use MCP when you want an AI client to query analytics, set up reports and dashboards or update your brand styling, without building your own API layer first.

Use the Cevoid API instead for direct REST integrations from your own application code. For the brand team's guide to connecting an AI assistant and what it can do in the platform, see MCP server in the Help Center.

The MCP server URL is https://mcp.cevoid.com/mcp. Use a client that supports MCP over streamable HTTP with OAuth, such as Claude, Cursor or ChatGPT.

1. Open an MCP client such as Claude, Cursor or ChatGPT
2. Point it at https://mcp.cevoid.com/mcp
3. Complete the OAuth flow and choose which workspaces and permissions to grant
4. Ask your client to list your workspaces with list-workspaces and find the tools you need with tool-search
5. The client calls a discovered tool with execute-tool and the schema tool-search returned. Independent reads can run together with execute-batch; writes always run one at a time

Ask your client which workspaces and tools your connection can use, for example:

Which Cevoid workspaces and tools can I access?

If the connection works, the client answers with your authorized workspaces and tool categories instead of a generic answer from the model alone.

The Cevoid MCP server uses OAuth 2.1. Access tokens are short-lived, bound to the Cevoid MCP endpoint and carry both the principal and the authorized workspaces. Clients keep sessions alive with rotating refresh tokens.

• Access is limited to the workspaces selected during authorization.
• Analytics reads require analytics:read, including the entity and profile labels used in analytics dimensions and filter searches. Report and dashboard writes require analytics:write.
• BrandKit reads, previews, font discovery and icon discovery require brandkit:read. Styling changes, color scheme changes, font uploads and imports, icon overrides and publishing require brandkit:write.
• The connect page lets you choose which permissions to grant, and discovery only returns the tools your connection allows. Reconnect an existing client to grant BrandKit permissions.

Use the discovery documents instead of hard-coding OAuth endpoints when your client supports discovery.

Request only the scopes your client needs, and send the MCP endpoint as the OAuth resource. Access tokens expire after 15 minutes. Refresh tokens rotate on every use and expire 30 days after the latest refresh; a client that is not used before then needs to reconnect through OAuth.

Tool calls are rate limited per authenticated principal and tool:

Calls through execute-tool or execute-batch also count toward the limit of the tool they invoke.

Your client finds domain tools with tool-search and calls them by their canonical ID through execute-tool. Search without a query to browse the categories.

Workspace IDs come from list-workspaces; reuse them unchanged. Report IDs use the public rpt_… format and dashboard IDs use dsh_…. Treat these IDs as opaque, and reuse the returned values for later get, run, update and dashboard panel operations.

Analytics filter values are a separate contract. A returned value is an opaque token, valid for that filter and authorized workspace, and its format may vary. Reuse it unchanged and never treat it as a resource ID. Profile filter discovery needs at least two search characters and returns at most 20 label and token pairs.

List tools return compact summaries for choosing an item; the matching get tools return complete public definitions. Result counts use returnedRowCount, the rows actually returned after limits. analytics.run-report.executedQuery records the execution settings you need to interpret the results of a saved report.

BrandKit tools edit a shared draft. Each change saves right away, updates open BrandKit editors and replaces unsaved edits there, but live widgets and emails keep the published BrandKit until the draft is published, with brandkit.publish or in the BrandKit editor. The tools tell clients to publish only when you ask for it.

• Publishing: pass the draft revision from brandkit.get or from the last draft edit. A stale revision is rejected; review the changed draft, then publish again.
• Previews: brandkit.preview renders up to three components per call as images, from the draft by default or from the published BrandKit. It does not edit or publish anything.
• Writes: settings, scheme and icon writes do not take a revision, so a later write can overwrite an earlier one. A write that collides with another draft save is rejected with draft_changed; read the draft again and retry. Omitted settings stay unchanged, and null resets a setting to its default or inherited value. Dimensions are in pixels, line height is a multiplier and letter spacing is in em.
• Fonts: for a local font, prepare the upload, send the file to the returned URL, then complete the registration. You can also import a public WOFF2, WOFF, TTF or OTF file of up to 10 MB; URL redirects are rejected. Assign the returned font ID to typography or buttons with brandkit.update.
• Icons: custom icons replace existing editable slots. Supply a complete monochrome SVG with a viewBox, up to 16 KiB per icon; active content and external resources are rejected. Icons inherit the theme color, and rating stars use the BrandKit star design setting. Reuse font and icon pack IDs from discovery unchanged.

These analytics resources require analytics:read:

• cevoid://analytics/query-schema
• cevoid://analytics/dimensions
• cevoid://analytics/filters
• cevoid://analytics/filter-values
• cevoid://analytics/reports
• cevoid://analytics/capabilities

The client discovers the relevant tool before it runs it.

• MCP server: connect an AI assistant to ask about analytics, build reports and dashboards and edit the BrandKit, explained for the brand's team.
• Cevoid API: REST integrations with scoped secret keys.