Elicitly
Human-in-the-loop for AI agents over MCP elicitation: confirm/form dialogs plus a capability doctor
Links
README
From the repo.
Elicitly — Free Edition
Your AI has questions. Your users have answers. Elicitly connects them — human-in-the-loop for prompts and Agent Skills over MCP elicitation.
This repository is the open-source Free Edition: the elicitly local MCP
server and the @elicitly/tools library. The hosted Pro Edition — durable
approvals with review pages, an audit trail, and organization dashboards —
lives at elicitly.ai.
Quick start
Run the local (stdio) MCP server straight from npm — no clone, no config:
npx -y elicitly
Register it in your MCP host (Claude Code shown; any stdio host works):
claude mcp add elicitly -- npx -y elicitly
Or skip the command line: in Claude Desktop, download elicitly.mcpb from
the latest release
and open it — a one-click, self-contained
desktop extension.
Then ask your agent to call elicit_confirm ("Ship it?"), elicit_form
(your own JSON schema), or elicit_doctor (does this host actually support
elicitation?).
Packages
| Package | What it is |
|---|---|
elicitly | The local stdio MCP server (npx -y elicitly) |
@elicitly/tools | Embeddable toolkit: register the elicitation tools on your own McpServer |
Why
MCP has a native answer for asking a human — elicitation — but host support is
inconsistent, and only an MCP server can trigger it. Nobody wants to roll
their own server just to ask a question. Elicitly is that server (or, via
@elicitly/tools, the toolkit inside yours), plus the diagnostic
(elicit_doctor) that makes real-world host support visible — feeding the
public Support Matrix.
Develop
pnpm install
pnpm build # tsdown builds both packages
pnpm test # vitest
pnpm test:coverage # vitest + v8 coverage (thresholds enforced; HTML report in coverage/)
pnpm typecheck # tsc
pnpm check # biome
License
Config for your environment
Replace {MCP_ENDPOINT_URL} with this MCP’s endpoint URL (from its repo or docs above). No API key — you connect directly.
Tool
OS
Config file: ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"url": "{MCP_ENDPOINT_URL}"
}
}
}Paste into mcpServers in the config file. Restart Cursor after saving.
If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.