How to Connect Cline to a YouTube MCP Server

Cline adds Prepublish as a remote MCP server, either from the Remote Servers tab in its MCP settings or as one entry in the extension's MCP settings JSON:

{
  "mcpServers": {
    "prepublish": {
      "type": "streamableHttp",
      "url": "https://mcp.prepublish.ai"
    }
  }
}

That type field is not decoration. Cline's documentation states that omitting it defaults to the legacy sse transport, and Prepublish serves Streamable HTTP, so an entry without the type asks for a transport that is not there. The shared reference, including the six tools and the published limits, is the Claude Code hub guide. This page covers what is specific to Cline.

Cline documents two routes in MCP, read on 2026-09-21.

The Remote Servers tab, which is the shortest path. In the Cline panel, click the MCP Servers icon, the stacked server icon in the top toolbar. Open the Configure tab. Then, instead of editing JSON, use the Remote Servers tab and fill in three fields: Server Name, which is any unique label, Server URL, which is https://mcp.prepublish.ai, and Transport Type, where you choose Streamable HTTP rather than SSE (Legacy). Click Add Server.

The settings JSON. In the same Configure tab, click Configure MCP Servers, which the documentation describes as the button near the bottom that opens the MCP settings JSON used by the extension. Add the entry shown above under mcpServers, save, and the server appears in the list.

One naming note. The page read on 2026-09-21 calls the extension's file the MCP settings JSON and gives ~/.cline/mcp.json as the CLI file. It does not name cline_mcp_settings.json, the filename that circulates in third-party guides, so that name was not verified on the vendor page on 2026-09-21. Use the button in the panel, which opens the file the extension actually reads.

Cline's CLI carries the same server definitions and its own configuration file at ~/.cline/mcp.json, per the same page.

The wizard handles it interactively. Run cline mcp and the menu can list servers and their enabled state, add a server, edit one, enable or disable one without deleting it, and delete one. Adding a server prompts for the name, the transport type, and then either a command and arguments for a local server or a URL and headers for a remote one.

The non-interactive route lists what is configured: cline config mcp prints the servers and cline config mcp --json prints them as JSON. That confirms the entry you added is the entry Cline is reading.

The page's security guidance applies here directly: keep secrets in environment variables, and limit autoApprove to tools you would let run unattended. Three of the six tools take an email address or a script you may not want auto-submitted, so leaving autoApprove empty is the conservative setting.

Five failures are Cline's rather than the server's.

The transport type defaults to SSE. A remote entry with a url and no type is read as the legacy sse transport, per the documentation. Prepublish answers a POST with one JSON response and no event stream, and a GET returns the server card rather than an SSE stream, so an SSE client has nothing to subscribe to. Set "type": "streamableHttp".

The server is disabled or was never restarted. Enable, disable and restart are separate controls, and a disabled server is not an error, it is absent. Restart the server from the MCP settings panel after a change.

The request times out on a full audit. Cline's troubleshooting table names timeout errors and the fix of increasing the MCP timeout. This one matters before you run audit_script: that tool starts an asynchronous backend job and polls for up to 90 seconds inside the same request, per prepublish-fe/lib/mcp/tools.ts. A shorter client timeout cuts the call off before the audit settles. The audit is not lost, because the tool returns an analysis_id and tells the model to collect it with get_audit, so the recovery is one more tool call.

No tools appear. The troubleshooting table gives the fix directly: confirm the server started and exposes tools. A reachable endpoint with a failed tool list is almost always the transport mismatch above.

An authentication error appears for an entry you did not think needed one. Only the optional pp_live_ key lives in headers, so an empty headers object is correct and a malformed Authorization header is the usual cause.

When the problem is on the server side, the response says what it is. These strings come from the endpoint's own code.

Cline supports both shapes, so the decision is yours rather than the client's.

Cline's examples show a local server as command plus args plus env, and a remote one as type plus url plus optional headers. Prepublish swaps which one you write:

{
  "mcpServers": {
    "prepublish": {
      "command": "npx",
      "args": ["-y", "prepublish-mcp"]
    }
  }
}

Use the hosted URL by default. It needs no Node installation, it is one line, and it keeps the tool list identical to every other client's.

Use the bridge when you want the process on your machine. It connects to the same hosted endpoint and re-serves it over stdio, and prepublish-mcp/test/bridge.test.ts asserts that its sorted tool names equal the six in the hosted list. It needs Node 20 or newer, per the engines field in prepublish-mcp/package.json, and PREPUBLISH_MCP_URL and PREPUBLISH_TOKEN point it elsewhere or authenticate it. On startup it writes one line to stderr naming the upstream URL, the tool count and whether a token was supplied, which is what Cline's server output shows if the process starts and then fails.

Ask for the free arithmetic first, because it proves the connection and costs nothing:

Use the prepublish server to tell me how many words I need for a twelve minute video, and how long a 2,400 word script would run.

The response carries speaking_rates_words_per_minute with fast, median and slow, a source string, a caveat string, word_count, runtime_minutes and word_budget, the last two each with the same three keys. The rates are the measured ones in prepublish-fe/lib/seo/wpm-data.ts. Computed from them, 2,400 words is 15 minutes at 160 words per minute, 13.3 at 181 and 11.9 at 201, and a twelve minute target needs 1,920 words at the slow rate, 2,172 at the median and 2,412 at the fast rate. The caveat field carries the limit in the payload: speaking time only, with pauses, B-roll, demonstrations and edits adding time.

Then hand over an opening, which is where a coding agent is comfortable:

Use the prepublish server to audit this hook and name the sentence that loses attention first: "Most creators lose half their viewers in the first fifteen seconds, and the reason is rarely the editing."

audit_hook takes hook_text, an optional niche and an optional email, and returns a hook object with evaluation_id, hook_text, niche, overall_score, grade, one_sentence_verdict, sentences, top_issues, rewrites, model_used and created_at. Each entry in sentences carries its own attention read, and each rewrite is written to be pasted over the line it replaces. This tool counts against the allowance.

The payload repeats its own limit, and the model relays it: "This is a text-only check of an unrecorded script. It maps relative attention risk inside the draft. It does not measure or predict published YouTube retention, and it cannot account for delivery, editing, thumbnail, topic or distribution."

The check reads text. It does not watch a video, fetch a transcript or read your channel, and no result is a forecast of audience behaviour.

The free tier is published in prepublish-fe/lib/mcp/limits.ts, and the values in force for the current day are in the server card at https://mcp.prepublish.ai.

A full audit is released by email, so audit_script asks for an address and the backend rejects a request without one. The narrower checks do not ask. An optional pp_live_ API key in the headers block takes the call out of the shared daily cap and applies your account's plan quota instead, and it is the reason Cline's own advice to keep secrets in environment variables is worth following here.