MCP server

The SpikeMe MCP server speaks the Model Context Protocol, so an agent like Claude Code or Cursor can call SpikeMe directly. Your agent already reads code and answers questions; the MCP server lets it produce the one artifact it otherwise can't: a grounded, time-boxed spike document.

Tools

The server exposes three tools:

• analyze_stack: analyze the repo's manifest (package.json, pyproject.toml, requirements.txt, go.mod), returning the stack, categories, and gaps that make good spike topics. Local and free, no account needed.
• generate_spike: generate a full spike decision document for a topic, grounded in the repo's real stack and live registry facts. Requires spikeme login and an active plan.
• generate_implementation: given a chosen library, generate an actionable implementation guide — the version-pinned install command (from live facts), a minimal working example in the repo's stack, timeboxed integration steps, a proof-of-concept checklist, and a rollback plan. SpikeMe returns the plan for the agent to run under your confirmation; it does not execute anything. Inputs: library (required), path, topic, language. Requires spikeme login and an active plan.

Setup

The MCP server is the spikeme-mcp package. Nothing to install globally — npx fetches and runs it on demand. For generate_spike and generate_implementation, sign in once with the CLI:

npx spikeme login        # needed for generate_spike and generate_implementation (active plan)

Claude Code

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

Cursor and other MCP clients

Add SpikeMe to your MCP config (for example .cursor/mcp.json or your client's mcpServers block):

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

If you prefer a global install, run npm install -g spikeme-mcp and use spikeme-mcp directly as the command.

Authentication

generate_spike and generate_implementation reuse the credentials from npx spikeme login (stored in ~/.spikeme/credentials.json), so sign in once in your terminal. They need an active plan; analyze_stack is always free. You can also set the SPIKEME_TOKEN environment variable in the server's env for headless setups.

Example prompts

Once configured, ask your agent things like:

• "Analyze this repo's stack and suggest a spike topic."
• "Generate a spike comparing TanStack Query and SWR for this project."
• "Spike whether we should adopt Zod or Valibot here, quick depth."
• "Now write me an implementation guide for the chosen library."

The agent calls analyze_stack, generate_spike, or generate_implementation, and the document comes back inline.