There are two ways to run SpikeMe: the web app, which needs no install, and the CLI, which runs from your terminal. Pick whichever fits where you work.
Web
No install. Go to spikeme.io/app and sign in with a
magic link sent to your email. Once you are in, upload a manifest
(package.json, pyproject.toml, requirements.txt, or go.mod), pick a topic,
then
read and share the generated document.
CLI
The CLI needs Node.js 18 or newer. Run it once without installing anything:
npx spikeme
Or install it globally so the spikeme command is on your PATH:
npm install -g spikeme
spikeme
Running spikeme with no arguments analyzes the current repo's stack. This step
is free and fully local: nothing is uploaded. It prints your stack, categories,
and the gaps that make good spike topics.
Coding agent (MCP)
If you work with an agent like Claude Code or Cursor, SpikeMe also runs as an MCP server — the agent analyzes your repo and generates the spike document without leaving the editor. No global install:
claude mcp add spikeme -- npx -y spikeme-mcp
Sign in
Generating documents (spikeme gen) runs through the SpikeMe backend, so it
needs an account. Sign in from the terminal:
spikeme login
This starts a device-flow login: the terminal shows a one-time code and opens
your browser to https://spikeme.io/cli-login, where you type that code and
approve it (you must be signed in to spikeme.io). Only enter a code from a login
you started yourself. The token is saved locally and refreshes on its own. See
the CLI reference for the full flow, CI tokens, and plan
requirements.
What's next
- Generating spikes: pick a topic, choose depth and options, and read how the evidence engine grounds the document.
- CLI reference: every command, flag, and environment variable.
- MCP server: use SpikeMe from inside Claude Code or Cursor.