Hunch, from your AI client.
The Hunch MCP server at https://hunchsheet.app/mcp puts five tools in Claude, ChatGPT, Gemini CLI, Cursor and VS Code: hunch_ask, hunch_pick, hunch_score, hunch_multi and hunch_balance. Each one judges a batch of short texts and returns a calibrated number: a probability, a pick from your own options, a score on your own scale, or several yes/no answers at once.
Same key, same credits, same six-hour cache as Sheets and Excel. No math, counting or dates: the model judges text.
Get a key
One key works everywhere: Sheets, Excel, the CSV runner and every client below.
100 free rows
The trial form below emails a key with 100 free rows. No card.
Buy more
5,000 rows for $29, 25,000 more for $19. Credits never expire. Buy a pack.
The key itself
hunch_ plus 40 hex characters. One credit per answered text per question. Blank texts and texts repeated in the same call are free.
Connect a client
Add to Claude or ChatGPT: paste https://hunchsheet.app/mcp and sign in with your key when prompted. No key to copy into a URL, nothing to keep secret in your client's settings file. Steps and field names below were checked against each vendor's own current docs on 2026-09-28; menus move between releases, but the URL and JSON forms here still work as written.
Claude (claude.ai and Claude Desktop)
Claude calls this a custom connector.
- Open Customize › Connectors at claude.ai (in the desktop app, Settings › Connectors), then Add custom connector.
- Paste this URL as the server URL, with no key in it.
https://hunchsheet.app/mcp- Click Add. Claude opens Hunch's sign-in page; paste your key there and click Connect. Claude gets a token back and never sees the key itself.
A free account can add one custom connector; on Team and Enterprise, an owner adds it once and members connect to it. If your version of Claude doesn't offer sign-in for a custom connector yet, use the URL-with-key fallback below instead, with No sign-in.
Claude Code
One command, run once from a terminal:
claude mcp add --transport http hunch https://hunchsheet.app/mcpClaude Code opens a browser to Hunch's sign-in page on first use; paste your key there once. claude mcp list shows what is connected. If your version doesn't open a browser automatically, use the URL-with-key fallback below instead.
Add --scope project to check the connection into a repository for a team, or --scope user to use it in every project.
ChatGPT
- In ChatGPT, open Settings, then Apps & Connectors, then Advanced settings, and turn on developer mode.
- Add a connector with this URL.
https://hunchsheet.app/mcp- ChatGPT opens Hunch's sign-in page; paste your key there and click Connect.
Developer mode is on Plus, Pro, Business, Enterprise and Edu plans. It is not on the free plan. If your version doesn't prompt you to sign in, use the URL-with-key fallback below instead, with no authentication.
Gemini CLI
Add a server to ~/.gemini/settings.json.
{
"mcpServers": {
"hunch": {
"httpUrl": "https://hunchsheet.app/mcp",
"headers": { "Authorization": "Bearer hunch_your_key_here" }
}
}
}The field is httpUrl, not url. That one is for the older SSE transport. Restart the CLI after saving.
Cursor
Add a server to ~/.cursor/mcp.json for every project, or .cursor/mcp.json in one project.
{
"mcpServers": {
"hunch": {
"url": "https://hunchsheet.app/mcp",
"headers": { "Authorization": "Bearer hunch_your_key_here" }
}
}
}Or add it with one click (it ships with a placeholder key, edit it in Cursor's MCP settings afterward):
VS Code
Run MCP: Add Server from the Command Palette, choose HTTP, and paste https://hunchsheet.app/mcp. Or add it directly to .vscode/mcp.json.
{
"servers": {
"hunch": {
"type": "http",
"url": "https://hunchsheet.app/mcp",
"headers": { "Authorization": "Bearer hunch_your_key_here" }
}
}
}VS Code names the top-level key servers. Cursor and Claude Code use mcpServers.
Or add it with one click. VS Code will prompt for the key itself:
URL-with-key fallback (no sign-in)
For a client that doesn't offer sign-in for a custom connector, or a config file that takes a header, put the key directly in the URL instead. This still works everywhere; it's the same server either way.
https://hunchsheet.app/mcp/hunch_your_key_hereIn Claude, choose No sign-in as the authentication method. In ChatGPT's connector form, choose no authentication. Claude Code: claude mcp add --transport http hunch https://hunchsheet.app/mcp/hunch_your_key_here.
Use it with a Google Sheets or Excel MCP server
Hunch does not read or edit your spreadsheet files. A Google Sheets MCP server, or an Excel one, reads and writes the cells. Hunch's tools judge the text in between. Run both in the same client and describe the whole job in one prompt.
Read the column
The spreadsheet MCP server's own read tool pulls column A into a list of texts.
Judge each text
The client calls a Hunch tool, say
hunch_pick, on that list with your categories or question.Write the labels back
The spreadsheet MCP server's write tool puts the results into column B.
One prompt covers all three steps, for example: "Read column A of Sheet1, use hunch_pick to sort each row into billing, refund, bug or other, then write the picks into column B."
Example prompts
- "Score these 20 support tickets for urgency with hunch_ask and list the top 5."
- "Use hunch_pick to sort these leads into decision maker, influencer or no fit."
- "Rate these 50 reviews from angry to delighted with hunch_score, then tell me the average."
- "For each of these emails, use hunch_multi to tell me if the sender can buy, is angry, and needs a reply today."
- "Check hunch_balance before you start. Stop and tell me if fewer than 100 credits are left."
What each tool costs
| Tool | Cost | Returns |
|---|---|---|
| hunch_ask | 1 / text | A probability, 0 to 1. |
| hunch_pick | 1 / text | A chosen option and its confidence. |
| hunch_score | 1 / text | A position on your scale, a label, a confidence. |
| hunch_multi | 1 / question / text | One probability per question. |
| hunch_balance | Free | Credits left. |
Blank texts and texts repeated elsewhere in the same call are free. Answers are cached for six hours, so asking again inside that window costs nothing.
Security
Signing in through a client's own "Connect" flow never puts your key in a URL: the client gets a separate token, and never sees the key itself. Disconnecting the client in its own settings revokes that token. If a client doesn't offer that, write to us and we revoke it from here; your key and its other connections are unaffected.
The URL https://hunchsheet.app/mcp/hunch_... is a secret. Anyone who has it can spend your credits. Treat it like a password: do not paste it into a public chat, a shared doc or a public repository.
When a client supports the Authorization-header form, that one is a little safer, since a header is less likely to end up in browser history or a screenshot than a URL.
If a key leaks, write to us and we rotate it. The old key stops answering. The new one carries the same balance.
Limits
- The model judges text. No math, counting or dates.
- English works best. Other languages work with lower accuracy.
- Put the full definition of what counts as yes, or of each option or level, inside the question. The model sees nothing else about your text.
- Up to 500 texts per tool call. Hunch chunks them internally.
hunch_multitakes 1 to 10 questions per call.hunch_picktakes 2 to 255 options;hunch_scoretakes 2 to 10 levels.