About Idea Launch
Idea Launch connects coding agents to the Idea Launch MCP server to set up ad tests for SaaS or app ideas and review market results from an editor. It also includes skills for preparing a test brief and interpreting results; users must start paid tests in the web app.
Key features
- 01Set up product ad tests
- 02Read clicks, signups, and validation scores
- 03Get recommended next steps
- 04Use skills for briefs and result interpretation
Use cases
- 01Assess a SaaS or app idea
- 02Test positioning for a shipped product
- 03Review test results in an editor
MCP support
The plugin connects to the Idea Launch MCP server, using browser sign-in without API key configuration. The server supports setting up ad tests and reading results; agents cannot start paid tests.
Configuration
Remote servers need no local install. Pick your client, then copy the config or install in one click.
https://www.idea-launch.io/api/mcpclaude mcp add --transport http io-idea-launch-idea-launch https://www.idea-launch.io/api/mcp{"mcpServers":{"io-idea-launch-idea-launch":{"type":"http","url":"https://www.idea-launch.io/api/mcp"}}}Open in Cursor{"servers":{"io-idea-launch-idea-launch":{"type":"http","url":"https://www.idea-launch.io/api/mcp"}}}Open in VS Code{"mcpServers":{"io-idea-launch-idea-launch":{"type":"http","url":"https://www.idea-launch.io/api/mcp"}}}{"mcpServers":{"io-idea-launch-idea-launch":{"serverUrl":"https://www.idea-launch.io/api/mcp"}}}Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.
Tools
Details
From the project READMEIdea Launch — Agent Plugin
Idea Launch runs a small, managed ad test for a SaaS or app idea, or for a product you already shipped, and returns a market read: click-through rate, signups, a validation score and a recommended next move. This plugin lets Claude Code, Cursor, Codex and other coding agents set up that test and read the result without leaving the editor. It connects the Idea Launch MCP server (browser sign-in, no API key) and adds two skills: one for writing a brief worth testing, one for reading the result honestly. No agent can spend money; you start each paid test in the web app.
In Claude Code:
/plugin marketplace add apsquared/idea-launch-plugin
/plugin install idea-launch@apsquared
The two skills only, for any agent that reads SKILL.md (connect the server as
well): npx skills add apsquared/idea-launch-plugin. Every other client, and
the MCP server on its own: https://www.idea-launch.io/mcp.
The package
A portable Agent Plugins v1.0.0 package that connects the Idea Launch MCP server and adds the judgement needed to use it well.
One package, two manifest formats. The Agent Plugins entrypoints sit at the
root; Claude Code reads its own manifests from .claude-plugin/ and .mcp.json.
Both formats read the same skills/ directory, unmodified.
agent-plugin/
├── plugin.json # Agent Plugins manifest
├── mcp.json # Agent Plugins MCP server (streamable-http)
├── .claude-plugin/
│ ├── plugin.json # Claude Code manifest
│ └── marketplace.json # Claude Code marketplace entry (source ./)
├── .mcp.json # Claude Code MCP server (http)
├── LICENSE # MIT
└── skills/
├── reading-a-market-read/SKILL.md # how to interpret a result honestly
└── setting-up-an-idea-test/SKILL.md # how to get a test worth paying for
The two MCP files describe the same server at the same URL and carry no
credentials. They differ only in the name each format gives the transport:
Agent Plugins calls it streamable-http, Claude Code calls it http.
Why a plugin and not just the MCP server
The MCP server exposes capability. It does not carry judgement, and that is the part that decides whether a test is any good:
- A
validationScoreof 45 at low confidence, on atraffic_backedread, is easy to misreport as "mixed signal, refine the angle". It is closer to "not settled, and it cannot speak to demand at all". - A
targetCustomerof "founders" produces targeting that underperforms, and the user blames the product rather than the brief.
Both skills also restate the two hard rules: an Ad Run is evidence about a positioning rather than proof about a business, and no agent can spend the user's money — starting a run is confirmed by the user in the web app.
Installing — Agent Plugins clients
Install through any client that supports the Agent Plugins format; the client handles discovery, installation, and permissions. Installing the plugin also configures the MCP server, so there is no separate connection step.
For a client that does not support the format yet, connect the MCP server directly — see https://www.idea-launch.io/mcp.
Installing — Claude Code
Add the marketplace, then install from it:
/plugin marketplace add apsquared/idea-launch-plugin
/plugin install idea-launch@apsquared
Then complete the sign-in:
/mcp
Pick idea-launch and authenticate. Until you do, the server answers 401 and
Claude Code reports it as failed to connect — that is the unauthenticated state,
not a broken install. The skills load either way; the tools need the sign-in.
Installing — Cursor
The button connects the MCP server only. Cursor also supports the Agent Plugins format, which adds the skills too — see https://cursor.com/docs/plugins.
Installing — skills only, any agent
For an agent that reads SKILL.md files but not this plugin format, install
the two skills with the skills.sh CLI:
npx skills add apsquared/idea-launch-plugin
That installs the skills, not the server. Connect the MCP server as well (https://www.idea-launch.io/mcp), since the skills call its tools.
Auth
None to configure. Neither mcp.json nor .mcp.json carries credentials by
design. The first call opens a browser sign-in over OAuth 2.1 with dynamic
client registration.
License
MIT — see LICENSE. The package is manifests and skill documents; the Idea Launch service itself is governed by its terms of service.
Conformance
This package targets Agent Plugins v1.0.0 and the Agent Skills specification,
and additionally ships the Claude Code plugin layout.
Conformance is enforced by an automated check on every change: the manifest's
closed field set and name rules, the streamable-http server shape, and each
skill's frontmatter — including that a skill's name matches its directory.
The check also asserts that mcp.json carries no credential-shaped values,
since this package is distributed publicly.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.