Sprites
Perplexity173d livePerplexity Computer
Add Sprites to Claude
Available on Free (one custom connector), Pro, Max, Team and Enterprise, on claude.ai, Claude Desktop and Claude Code. You need a Sprites account with your channels already connected — the connector acts on that connection, so Claude never handles a platform credential.
Open the connector settings
On a Free, Pro or Max plan, go to Settings, then Customize, then Connectors. On Team and Enterprise an Owner adds the connector once for the whole workspace under Organization settings → Connectors, and everyone else only clicks Connect.
ClaudeSettingsCustomizeConnectorsAdd a custom connector
Click the + button, then Add custom connector. On Team and Enterprise, hover Custom and choose Web. Paste the Sprites endpoint and click Add. Leave Advanced settings alone — Sprites publishes its own OAuth client metadata, so there is no client ID or secret to fill in.
Add custom connectorNameSpritesRemote MCP server URLhttps://mcp.sprites.ai/mcpSign in and approve the scopes
Click Connect. Claude opens the Sprites consent screen in a new tab. Sign in with the account your ad platforms are already connected to, review what Sprites is being allowed to do, and approve. The grant is revocable from Claude settings or from Sprites at any time, and the very next tool call after a revoke fails closed.
Spriteswants access on behalf of Claude- Read your connected ad accounts and site data
- Draft campaigns, creatives and pages for your approval
- Apply only the changes you approve
AllowCancelOr add it to Claude Code in one line
The terminal client takes the same endpoint. Run the command, then type /mcp inside Claude Code to finish the browser sign-in.
Terminalclaude mcp add --transport http sprites https://mcp.sprites.ai/mcp
Which Claude surface you are on
Only the surfaces below are documented here. Where we have not verified a path, the page says so rather than implying the setup is the same everywhere.
claude.aiVerified
Settings → Customize → Connectors. On Team and Enterprise an Owner adds it once for the workspace.
Claude DesktopVerified
The same connector list as claude.ai, signed in to the same account.
Claude CodeVerified
One command: claude mcp add --transport http, then /mcp to finish the browser sign-in.
What Claude can reach
All 15 channels run through the one connector. Rows with a guide go deeper on that pairing; the rest are covered on the channel's own page.
| Channel | What Claude does with it | Guide |
|---|---|---|
| Google Ads | Claude draws the approval card inline, so a budget or a negative-keyword list is edited in the thread and applied on click. | Set up → |
| Meta | Creative comes back as a preview in the real placement rather than as a description of one. | Set up → |
| Cost per lead broken down by job function and seniority, which is the cut Campaign Manager makes hardest. | Set up → | |
| TikTok | Creative fatigue and hook performance summarised in prose, which is the format this channel’s data is least readable in natively. | Set up → |
| Subreddit-level performance beside what the comments actually say, since Claude will read both in one turn. | Set up → | |
| GA4 | Ask in English; Claude picks the dimensions, runs the report and states what it queried so the working can be checked. | Set up → |
| PostHog | Funnels, retention and trends by campaign, read-only, beside the spend that bought them. | Set up → |
| SEO | Posts land in your CMS as drafts and page edits arrive as a diff, both inside the same conversation. | Set up → |
| AEO | Ask which prompts you are cited in and Claude reports the answer engines that mention you, including its own. | Set up → |
| Shopify | Product performance and ad spend in one question, with the store read live rather than from an export you made on Tuesday. | Set up → |
| GBP | Review responses drafted in your voice and held as approvals, rather than posted the moment they are written. | Set up → |
| WordPress | A researched post arrives as a WordPress draft with its metadata and internal links set — the one workflow where Claude’s long-document handling is the reason to use Claude specifically. | Set up → |
| Webflow | Researched CMS posts, page metadata and copy edits, each an approval before it applies. | Set up → |
| Framer | Researched posts and SEO fields written into your Framer CMS, added after you approve. | Set up → |
| Zoho CRM | Pipeline and lead-source reporting beside ad spend, with approved lead, deal and note writes. | Set up → |
Marketing workflows worth doing in Claude
The only host that draws the Sprites approval card in the chat, so you edit a budget or a headline in place and approve it without leaving the thread.
Cross-channel performance briefing
One question covering Google, Meta and LinkedIn together, with the period comparison and the one number that actually moved — not three dashboards you reconcile by hand.
Search-term triage
The queries that spent without converting, ranked by waste, and the negatives that would have stopped them — proposed as one approval rather than forty clicks.
Creative fatigue diagnosis
Frequency and CTR trend per Meta creative, so the ad that broke is named rather than the campaign it sits in.
Landing-page investigation
GA4 entry pages against conversions, to find the page taking a third of the traffic and converting at nothing.
SEO opportunity plan
Search Console queries ranking 8-20, the ones within reach of the top five, and a post drafted into the CMS for the best of them.
Permissions, approvals and authentication
You sign in with OAuth and the connector acts on your workspace. No API keys in a config file, no shared service account, and revoking the grant kills the very next call.
| Action | When it happens | What that covers |
|---|---|---|
| Read an account | On request | Spend, performance, search terms, rankings, catalogue |
| Draft a change | On request | Campaigns, creative, budgets, pages and posts |
| Apply a change | Only on your approval | Claude draws the approval card in the thread; it applies when you click |
| Spend money | Only on your approval | New campaigns start paused unless you say otherwise |
| Revoke access | Instantly | The next call after a revoke fails, not the one after that |
Claude MCP questions
Which Claude plans support custom connectors?
Free, Pro, Max, Team and Enterprise. Free is limited to one custom connector. On Team and Enterprise an Owner adds it once under Organization settings → Connectors and everyone else clicks Connect.
Is Claude Code set up differently from claude.ai?
Same endpoint, different command. Claude Code takes it in one line with claude mcp add --transport http; the web and desktop apps take it through Settings → Customize → Connectors.
Can it change my ad accounts on its own?
No. Reads happen on request; every write comes back as a change you approve first, and new campaigns start paused. Nothing is applied because a model decided it should be.
Does the connector run on my machine?
No. It is a remote MCP server, so claude.ai reaches it from Anthropic's cloud. There is nothing to install and nothing to keep running.
What happens when I revoke access?
The very next tool call fails. Tokens are opaque and checked on each call rather than being self-describing, which is what makes a revoke immediate instead of eventual.
What is Claude MCP?
MCP is the Model Context Protocol, an open standard Anthropic published for connecting an AI assistant to external tools. "Claude MCP" is the shorthand for using it in Claude: you add a remote MCP server as a custom connector, and Claude gains tools it did not ship with.
Is a Claude connector the same as an MCP server?
Yes. Connector is Anthropic’s product word for a remote MCP server Claude can reach over HTTPS. Everything in the connectors directory is one, and any remote MCP server can be added by URL whether or not it is listed there.
How many MCP servers should I connect to Claude?
Three or four. Every connected server adds its tool descriptions to every turn, so past a few dozen tools Claude starts choosing badly and every message gets slower. The symptom looks like the model being careless; it is usually the setup being crowded.
How do I add an MCP server to Claude Code?
One command: claude mcp add --transport http <name> <url>, then /mcp inside Claude Code to finish the browser sign-in. It is the same endpoint the web and desktop apps take, so a server added in one place needs no separate configuration in the other.
Are Claude MCP connectors safe?
Treat adding one like installing an app with account access. Anthropic’s own documentation advises connecting only to trusted servers and warns that malicious MCP servers may include hidden instructions, because Claude reads tool descriptions as guidance. Check who publishes the server, what scopes it asks for, and whether writes are gated behind something you click.
Background worth having first
What the protocol is, how the servers differ, and what separates an agent from an assistant with tools attached.
- What is MCPThe protocol in plain English: what a server is, and what the latest revision changed.
- MCP vs APIThey are layers, not rivals. The one property that separates them.
- Best MCP servers for marketersTwelve that do marketing work, sorted by job — remote or local, read or write.
- Best Claude connectors for marketersEvery endpoint called rather than copied from a directory.
- AI marketing agentsWhat they are, what they automate, and how to evaluate one.
- AI marketing agent vs ChatGPTFive things ChatGPT cannot do, and the one that changed recently.
- Connect Muse to your ad accountsMeta’s agent has no connector for any ad platform. What it takes to give it one.
Point Claude at your ad accounts
One URL, one sign-in, and the next question you ask Claude about your marketing gets answered from the live account.