Agent guides

Use AfterLaunch with Claude on the web

Claude on the web, at claude.ai, reaches a remote MCP server as a custom connector, so a chat can read your ranked Feed and act on it from your account. Anthropic documents custom connectors on every plan, so the connector is the primary path and Claude Code with a key is the terminal-side alternative.

Connect

Connect it like this:

  1. Open claude.ai, go to Customize, then Connectors, and add a custom connector. On a Team or Enterprise plan an owner adds it once from the organisation settings instead.
  2. Paste the server URL: https://afterlaunch.io/api/mcp
  3. Then open Settings, Connectors, AfterLaunch and click Connect before you ask anything. Adding the connector does not sign you in, and a question asked first is answered without the tools.
  4. Approve the AfterLaunch consent screen, then ask Claude to read your AfterLaunch feed. The first tool call asks your permission.

Connectors are account wide, so once it is connected the tools are available in any chat rather than per project. Anthropic limits a free account to one custom connector.

No key yet? Read the agents page or hand your agent /llms.txt, and it can run the onboarding flow: it submits your site and email, you approve with one click by email, and nothing runs until you do.

What Claude on the web can do

The 55 tools are the same whichever agent calls them. Once connected, Claude on the web can:

  • Read the ranked Feed of growth moves, each with its full draft body, plus the product run, its automated loops and every drafted output. All read-only and free.
  • Act on a move: ship it, skip it with structured feedback, redraft the body before it goes out, or archive it. Free, on the act scope every self-serve key carries. Shipping stays the founder's decision; reporting it back the moment the work is live is the agent's, and it is what the engine measures the next ranking against.
  • Run the engine on demand: refresh an AI Visibility scan, top up the Feed, or run a single loop. These spend money, so they run on the write scope every key already carries and are metered: charged once per logical event, replayed on a repeat, and capped per tenant per day and per month.
  • Set the dials: a loop autonomy rung and its cadence. Both need the config scope and Founder tier, and social channels never accept full autonomy, so publishing stays review-and-post by a human.
  • Mint a Founder checkout link when the founder wants to upgrade. The link is always bound to the product run the key owns.
  • Write the Growth section into the repository it works in, the same block shown below, so the setup teaches the next session too.

In practice

With the connector added and connected, you ask Claude to triage growth work.

You

Read my AfterLaunch Feed and tell me the single highest-leverage thing to ship this week.

Claude on the web

Reading list_feed and get_visibility. The top item is an entity page so AI engines can resolve who you are; you are absent from Google AI Overviews today. Want the draft?

You

Yes, show me the copy before anything ships.

Claude on the web

Here it is from update_draft. Nothing ships until you approve it.

Illustrative. The tool names in brackets are the real read and act tools the agent calls.

Good to know for Claude on the web

  • Adding the connector is not the same as connecting it. Claude accepts it without a sign-in, because discovery on this server is deliberately anonymous, and then answers your first question without the tools until you have clicked Connect in Settings.
  • Connectors are account-wide, so once connected the AfterLaunch tools are available in any chat rather than per project, and in the Claude mobile app.
  • Anthropic documents custom connectors on Free, Pro, Max, Team and Enterprise, with a free account limited to one.
  • Prefer the terminal? Claude Code shares your Anthropic account and reaches the same 55 tools with a key.

Honest posture

Claude on the web operates the tools from a chat, not from your codebase, so there is no repository file to write. The remote server means nothing runs on your machine.

The AfterLaunch MCP server is remote and bearer-authenticated. There is no local process to install and nothing to execute on your machine: a config only points the agent at an HTTPS endpoint and attaches your key.

A key you mint yourself, or one an agent obtains through onboarding, carries the read, act and write scopes. Reading the Feed and shipping, skipping, redrafting or archiving moves work straight away. The metered pulls run on that same write scope, bounded by the product run credits and the per-tenant caps, and the autonomy dials need the config scope and Founder tier. Access opens in waves, so a refused key means your wave has not opened yet.

AfterLaunch drafts. A human approves and posts. It never claims to have posted to a social channel on its own.

Reporting back is part of the loop, not an extra step. When the founder confirms a move is live the agent calls ship_move, and when they decide against it the agent calls skip_move with the real reason. AfterLaunch measures what happens after a move ships, so an unreported ship is work the record never sees.

Questions

How do I connect claude.ai to a growth marketing agent?

Add https://afterlaunch.io/api/mcp as a custom connector under Customize, Connectors, then open Settings, Connectors and click Connect to sign in. Do that before your first question: the connector is added unauthenticated, and a question asked before you connect is answered without the tools.

Do I need an API key to use AfterLaunch from claude.ai?

No. The web connector signs you in with your AfterLaunch account, so there is no key to paste. Claude Code in the terminal uses a key you mint in AfterLaunch Settings, and both reach the same tools.

Is AfterLaunch a growth marketing MCP server?

Yes, though the server is the door rather than the product. AfterLaunch is the agentic growth engine: it watches how the leading AI search engines answer about you, audits every page of your site, tracks your competitors and the communities and directories that decide whether you get found, and keeps the history. The remote MCP server at https://afterlaunch.io/api/mcp speaks Streamable HTTP and exposes 55 tools, so an agent reads the ranked Feed of prepared growth work, ships or redrafts it, and reads the product run, its automated loops and its drafted outputs. Reads are free. The on-demand runs are metered and capped per tenant per day and per month.

Can I use it as an SEO MCP server or an AI visibility MCP?

The same server exposes get_seo, the site-wide SEO audit with search ranks, links and ready-to-apply fixes, and get_visibility, how ChatGPT, Gemini, Perplexity and Google AI Overviews answer questions about your product, with share of voice and cited sources. Both reads are free and need only a valid key.

Teach Claude on the web to use it

Claude on the web works from a chat rather than a repository, so there is no instructions file to write and nothing to install. The connector above is the whole setup. If you also work in a terminal or an editor, the guides below add the Growth section to that repository, so both surfaces reach for the same tools.

Or read the Growth section in your browser.

Using a different client?

Browse all agent guides, see the agents overview, or run a free growth tool first.