elicitly

Local (stdio) MCP server that gives your prompts and Agent Skills MCP elicitation — human-in-the-loop dialogs (elicit_confirm, elicit_form) and the elicit_doctor capability report, with no server of your own.

latest

Source

npm

Version: 0.7.2

Version published: 2 weeks ago

Weekly downloads: 62

Maintainers: 1

Weekly downloads

Created: 2 months ago

Source

elicitly — the Free Edition local MCP server

A local (stdio) MCP server that makes MCP elicitation easy to use from any prompt or Agent Skill — and its cross-host support visible.

Tools

elicit_confirm — an OK/Cancel confirmation modeled on JavaScript's confirm(); override labels.ok/labels.cancel to relabel or localize (e.g. Aceptar/Cancelar).
elicit_form — fire a form elicitation with your own JSON schema (requestedSchema subset); returns the raw {action, content}.
elicit_doctor — reports the connected host's support for the client features (elicitation form vs url mode, sampling, roots), and with probeElicitation: true runs one live elicitation round-trip and classifies the result (working, advertised_but_autocanceled, user_declined, advertised_but_unanswered, unsupported).

Run it

npx -y elicitly

claude mcp add elicitly -- npx -y elicitly

or in a host config file (.mcp.json or client settings):

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

Claude Desktop (one-click)

Download elicitly.mcpb from the latest release and open it in Claude Desktop (or Settings → Extensions → Install extension). The bundle is self-contained — no Node.js or npm setup of your own; it runs on Claude Desktop's managed runtime. Build it from source with pnpm --filter elicitly build:mcpb.

Claude Desktop and dialogs: current Claude Desktop builds don't advertise elicitation to local MCP servers (config-file and extension installs alike — measured, not assumed), so elicit_confirm and elicit_form return their graceful no-answer contract there rather than raising a dialog. elicit_doctor still reports the host truthfully — that report is the point. See the Elicitation Support Matrix for hosts where dialogs work, and the Pro Edition for approvals that reach a browser-based review page when the host has no native dialog.

Exercise it

elicit_doctor — passive capability report, no user prompt. Check support.elicitationForm / support.elicitationUrl.
elicit_doctor with probeElicitation: true — one real elicitation round-trip; read probes.elicitationForm.verdict.
elicit_confirm — an OK/Cancel prompt (e.g. message: "Ship it?").
elicit_form — your own JSON schema; inspect the raw {action, content}.

Full docs: elicitly.ai/docs

From source

pnpm install
pnpm --filter elicitly build   # produces packages/elicitly/dist/cli.mjs
claude mcp add elicitly -- node /absolute/path/to/packages/elicitly/dist/cli.mjs

Privacy

The server runs entirely on your machine: tool calls and elicitation answers stay between your MCP host and the local process, and nothing is sent anywhere. The contribute-fingerprint prompt only prepares a GitHub issue link — you review and submit it yourself. Project privacy policy: https://www.elicitly.ai/privacy/