Sensei MCP

Sensei is the operating system for solo founders — the smartest person in the room when the build stops being the hard part and the question becomes now what? This MCP server gives any AI coding tool (Claude Code, Cursor, Windsurf, Bolt, Continue, anything MCP-compatible) a read_product tool that returns Sensei's verdict on any URL — your competitor, your friend's startup, your own product. Free. One read per domain per day.

師 Sensei read linear.app
Score: 88/100  ·  Verdict: SHIP
A category-defining product for a market that already chose it.
…

What you get

• read_product(url) — Sensei reads the site cold and returns:
  • score (1–100) and verdict (ship / iterate / pivot)
  • oneLiner — the punchy version
  • bottomLine — 2-3 sentences in Sensei's voice
  • senseiObserves — the non-obvious thing the founder probably missed
  • costOfIgnoring — one line on the real consequence of doing nothing
  • marketAnalysis — summary, size, timing, target audience
  • competitionScan — summary + direct competitors
  • oneMove — gated. The Move beat unlocks at asksensei.dev.

Install

Claude Code

Add to your ~/.config/claude-code/mcp.json (or claude_desktop_config.json):

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

Cursor

Add to your ~/.cursor/mcp.json (project or global):

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

Windsurf

Add to your Windsurf MCP config:

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

Any other MCP-compatible tool

The server speaks stdio. Wire it up however your tool expects — the binary is sensei-mcp, no flags required.

Use it

From any of those tools, just ask:

read this product: https://linear.app

The agent will invoke read_product and surface Sensei's full reading. You can paste your competitor's URL, a Show HN post, your own staging site — Sensei reads what's on the page and tells you the truth.

Example tool call

read_product("https://linear.app")

Example response shape

{
  "url": "https://linear.app",
  "score": 88,
  "verdict": "ship",
  "oneLiner": "Category-defining velocity software with a buyer that already chose them.",
  "bottomLine": "Linear isn't competing on features — they're competing on feel...",
  "senseiObserves": "The pricing page leads with team size, not seats — that's a buyer signal.",
  "costOfIgnoring": "Every quarter you ship on Jira is a quarter your best engineers...",
  "marketAnalysis": {
    "summary": "...",
    "marketSize": "$2-4B project-mgmt SaaS",
    "timing": "Post-Jira fatigue, pre-AI-native disruption",
    "targetAudience": "Engineering leads at 10-200 person product orgs"
  },
  "competitionScan": {
    "summary": "...",
    "directCompetitors": [
      { "name": "Jira", "description": "..." },
      { "name": "Height", "description": "..." }
    ]
  },
  "oneMove": {
    "teaser": "Take a seat at asksensei.dev to see what to ship next.",
    "unlockAt": "https://asksensei.dev"
  },
  "stages": ["Reading the site", "Weighing the market", "Scanning the field", "Setting the verdict"],
  "blindReadId": "..."
}

Configuration

Both env vars are optional. The server points at production by default.

| Env var | Default | Notes | | ---------------- | ------------------------ | -------------------------------------------------- | | SENSEI_API_URL | https://asksensei.dev | Override for self-hosted or local Sensei. | | SENSEI_API_KEY | (unset) | Reserved for v2 authenticated tools. Not used yet. |

Rate limit

One blind read per domain per IP per 24 hours. If you hit the limit, Sensei says so plainly:

師 Sensei reads one domain per day. Try again tomorrow.

For unlimited readings on your own product — and the rest of Sensei (weekly readings, the Move beat, prediction tracking, decision memory) — take a seat at asksensei.dev.

Where this is going

v1 is the anonymous blind read — the same surface that lives at asksensei.dev/try, now reachable from inside your editor. v2 adds authenticated tools so Sensei becomes a working collaborator on your active product:

• get_my_active_reading — load the current week's Market Fit, Strategy, Presence, and Edge reads for the product you registered with Sensei.
• what_should_i_ship_next — pulls this week's One Move with the pattern note and the cost of ignoring.
• log_decision — close the loop. Tell Sensei what you shipped (or what you pushed back on) and it lands in your memories the next time you open the dashboard.

Same MCP server, opt-in SENSEI_API_KEY for the authenticated calls. Founders who've taken a seat will be able to do the entire weekly read without leaving Cursor.

Build from source

git clone <repo>
cd mcp-server
npm install
npm run build
node ./dist/index.js

License

MIT.