spikeme

Decide your next library with evidence, straight from your repo. spikeme reads your project manifest, maps your stack, and points at the decisions worth a time-boxed spike. Then it generates a full spike document (options, comparison, recommendation, PoC plan) grounded in live registry facts.

Supports npm (package.json) and Python (pyproject.toml, requirements.txt). More ecosystems are pulled in by demand.

Install

# one-off, no install
npx spikeme

# or install globally
npm install -g spikeme
spikeme --help

Quick start

# Analyze the current repo's stack — free, runs locally, nothing uploaded
spikeme

# Generate a full spike doc for a decision (needs a free SpikeMe account)
spikeme login                       # opens your browser to sign in
spikeme gen "state management library"

Commands

| Command | What it does | | --- | --- | | spikeme | Analyze the repo: stack, categories, and gaps that make good spike topics. Local and open. | | spikeme gen "<topic>" | Generate a full spike document for a decision. Runs through the SpikeMe backend (plan quota applies). | | spikeme login | Save your SpikeMe access token to ~/.spikeme/token. |

gen options

| Flag | Default | | | --- | --- | --- | | --depth quick\|full | full | Document depth | | --options 2\|3\|4 | 3 | How many options to compare | | --lang en\|pt | en | Document language | | -o, --out <file> | stdout | Write the doc to a file | | --path <dir> | current dir | Repo directory to analyze |

MCP server

SpikeMe also ships an MCP server — spikeme-mcp — exposing analyze_stack and generate_spike to coding agents (Claude Code, Cursor). Add it with (no global install needed):

claude mcp add spikeme -- npx -y spikeme-mcp

See the MCP docs for Cursor and other clients.

How it works

The analysis is local and open: parsing your manifest, categorizing dependencies, and spotting gaps all happen on your machine, with no upload. Only generation calls the SpikeMe backend, which runs the same evidence engine as spikeme.io: it collects live facts (versions, weekly downloads, licenses) from the npm registry or PyPI and grounds the document in real, dated numbers.

Auth

gen runs through the SpikeMe backend and needs an active plan (spikeme.io/precos) plus a signed-in CLI. Analysis (spikeme) is always free and needs no account.

spikeme login uses a device flow: it shows a one-time code, opens your browser, and you enter that code to approve. The token is stored in ~/.spikeme/credentials.json (mode 0600) and refreshes automatically. Only enter a code from a spikeme login you started yourself.

For CI or headless machines, pass a token directly:

spikeme login --token "<access-token>"   # non-interactive
# or
SPIKEME_TOKEN="<access-token>" spikeme gen "..."

License

MIT