Human-in-the-loop · MCP elicitation

Your AI has questions.Your users have answers.

Elicitly connects them: confirm, choose, fill a form, approve — from any prompt or Agent Skill, with no server of your own.

npx -y elicitly Read the documentation

elicitly wants to ask

Deploy v2.1.0 to production?

Cancel

DismissSubmit

elicit_confirm · rendered natively by your MCP host

→ { "confirmed": true }

How it works

1 — your skill asks

One tool call

No UI code, no server to write.

elicit_confirm({ message: "Deploy v2.1.0 to production?" })

2 — your user answers

A native dialog

Their MCP host renders it — the same host they're already working in.

3 — your skill continues

A typed result

Branch on the answer and carry on.

{ "confirmed": true }

From a real skill

# SKILL.md — deploy (excerpt)

Before deploying, call the elicit_confirm tool:

    message: "Deploy ${version} to production?"

Proceed only when the result is { "confirmed": true }.
On false or null, stop and report to the user.

Advertised ≠ working

Claude Code · stdio — form ✓ · url —Claude.ai (Cowork) · HTTP — form ✗ · url —Claude Desktop · HTTP — form ✗ · url —Claude Code · HTTP — form ✓ · url —ChatGPT Desktop · HTTP — form ✓ · url ✓ChatGPT Desktop · stdio — form ✓ · url —See the Support Matrix →

This gap is why the Pro Edition exists: when the host can't render a dialog, Pro's hosted review page works anywhere — your skill makes the same calls either way. How Pro adapts →

Editions

Free is the dialog; Pro is the durable process — approvals that survive disconnects, reach any device, and leave an audit trail.

Free

Runs entirely on your machine — no account, no service, no infrastructure.

npx -y elicitly

Pro

Free in Early Access

Hosted approvals with review pages, an audit trail, and TTL-bounded decisions that survive disconnects.

See pricing

Enterprise

Tell us more

Self-host in your cloud, SSO/SAML, compliance.