Tailwind's MCP (Model Context Protocol) server lets AI assistants securely access your Tailwind account. The server is located at mcp.tailwind.ai. You can connect in two ways: OAuth (enter the server URL in an assistant's built-in connector and authorize access) or an API key created under Settings → Integrations for CLI-based or config-file-based tools.
Two ways to connect
OAuth — used by AI assistants that have a built-in connector interface, such as Claude Desktop and Claude.ai (web/mobile) via Settings → Connectors → Add custom connector. You simply provide the server URL (mcp.tailwind.ai) and authorize access. There's no API key to paste.
API keys — used with CLI-based or config-file-based MCP setups (for example, manual mcp-remote bridges). Create a key under Settings → Integrations, then add it to your tool's MCP configuration.
Which method should I use?
If your AI assistant offers a connector with its own setup screen (including Claude Desktop and Claude.ai), use the OAuth method and just enter the server URL — you do not need to create or paste an API key. Only use an API key when your tool asks for one directly (typically CLI or config-file-based tools).
"Client registration pending approval"
This message means the OAuth app you're connecting hasn't been approved yet. Once approval completes, you'll be able to finish connecting.
"invalid_client" / "Client ID metadata document is malformed"
If connecting Tailwind from Claude.ai (web, desktop, or mobile) failed with {"error":"invalid_client","error_description":"Client ID metadata document is malformed"}, this was a bug on Tailwind's side — not a problem with your account, your plan, or your assistant.
Tailwind's MCP OAuth server validated the client metadata document that Claude publishes against a strict allow-list. When Anthropic added a new grant type to that document, Tailwind rejected the whole document and refused the connection. Connections from Claude Code were unaffected. Because the failure was server-side, clearing your browser cache, reinstalling, or reconnecting from a different device would not have helped.
This was root-caused and fixed on 31 August 2026. Tailwind now accepts grant types it doesn't recognise and simply ignores the ones it doesn't issue. If you hit this error, just try connecting Tailwind again — nothing needs to change on your end. If it still fails, email us at [email protected] and paste the error text exactly as it appears.
I'm connected, but my AI assistant says it has no access to my Tailwind tools
This is a common first-time snag, especially with Claude.ai's web and mobile apps. In most cases the connector is authorized and working fine — your assistant simply hasn't loaded its tools into that conversation yet. Work through these in order. The first step costs nothing and tells you whether you have a tool-loading problem or a genuine connection problem before you change any settings.
Ask for Tailwind by name. Claude.ai defaults to a Load tools when needed mode, which holds a connector's tools back until Claude goes looking for them. That means Claude can honestly report that it has no Tailwind tools while your connector is connected and perfectly healthy. Naming Tailwind in the request — for example, "use Tailwind to list my Pinterest boards" — prompts Claude to search for and load them. If that works, nothing was ever broken. You'll find the same steps in-product: in Tailwind's connect panel, open Set up or troubleshoot a client, choose the Claude.ai tab, and expand Not seeing your tools?
Turn the connector on in the chat itself. Adding Tailwind under Settings → Connectors only authorizes it — it does not automatically arm it in any given conversation. In Claude.ai, open the composer's tools menu (sometimes labeled "Search and tools") and toggle Tailwind on for that chat. This step is needed in every new conversation.
Check that individual tools are enabled. Under Settings → Connectors → Tailwind, each tool (Create Pinterest post, List Pinterest boards, etc.) has its own on/off switch. If all of them are off, your assistant will report it has no usable tools even though the connector shows Connected.
Confirm your plan supports custom connectors. On Claude.ai, custom connectors require a Pro, Max, Team, or Enterprise plan. On Team/Enterprise, an admin allow-list can also block a connector even after you've authorized it — check with your workspace admin if you're on one of those plans.
Restart your AI assistant. If you just connected Tailwind, fully closing and reopening the app (or starting a brand-new chat) can resolve tools not showing up.
Last resort: try the same server URL in a desktop or CLI client. If you've genuinely worked through everything above and the tools still don't appear, adding the same server URL (mcp.tailwind.ai) as a connector in Claude Desktop, or as an MCP server in Claude Code, will get you moving — your Tailwind authorization is the same, so there's nothing extra to set up on our end. Treat this as the last step rather than the first: switching clients when the real problem was tool loading moves a working connection for no reason.
If you've checked all of the above and tools still aren't available in the conversation, contact us at [email protected] and let us know which AI assistant you're using.
Connected — now what? Try the starter prompts
Once a chat client such as Claude.ai, Claude Desktop, or ChatGPT shows as connected, Tailwind's connect panel (Settings → Integrations) shows a "Now try asking your assistant" block directly in the connected state. It contains four copy-and-paste starter prompts arranged as a workflow — list your connected accounts, generate Pins from a URL, schedule them, then review your queue — and each prompt has its own copy button.
If you connected successfully but weren't sure what to ask next, start there.
Your plan's scheduling limit applies to Pins scheduled through MCP
Pins scheduled by an AI assistant (or through Tailwind's Public API) count toward your plan's post limit in the same way as Pins scheduled in the Tailwind web app. If a request would take you past your plan's combined published + scheduled limit for the period the Pin is scheduled into, Tailwind refuses it with an HTTP 402 (Payment Required) response, and your assistant will report that the Pin could not be scheduled. On plans with a fixed post limit the machine-readable reason returned is legacy_post_limit_exceeded; on credit-based plans you'll see the usual credit or trial denial instead.
A few details worth knowing:
Creating a draft does not count against the limit — only scheduling does.
Rescheduling a Pin that is already queued is not re-checked against the limit.
Unlimited plans have no cap to hit here.
If you're running into this limit, you can upgrade from your Tailwind Upgrade Page.
If you have more than one Pinterest account connected
Every Tailwind tool that acts on a specific account needs to know which account you mean, and your AI assistant chooses that account before the request ever reaches Tailwind. If you have several Pinterest accounts connected, the assistant is now prompted to ask you which one to use rather than picking one silently — but that prompt is guidance to the assistant, not something Tailwind can enforce.
The safest habit is to name the account in your prompt whenever it matters — for example, "schedule this Pin to my Home Decor account." If you only have one Pinterest account connected, nothing changes and you don't need to mention it.
More resources
Developers can find additional reference material in our developer documentation at api-docs.tailwind.ai/mcp, including a "Server connects but no tools appear" entry at api-docs.tailwind.ai/mcp/setup.