# elicitly [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/elicitly/elicitly  
**GitHub Stars:** 1  
**npm Downloads (last month):** 1014  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/elicitly

## Description
Human-in-the-loop for AI agents over MCP elicitation: confirm/form dialogs plus a capability doctor

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

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

## Documentation

## What elicitly MCP server does

The elicitly MCP server adds a human-in-the-loop step to MCP-based agents. It exposes tools for asking a person to confirm an action, collecting structured form data, and checking whether the current host can handle MCP elicitation. This is useful for workflows where an agent should pause before proceeding or needs information that a user must provide.

The project is distributed as the Free Edition. It includes a local stdio server named `elicitly` and a separate `@elicitly/tools` package for developers who want to register the tools on their own `McpServer`. The hosted Pro Edition is a separate service and is not part of this local package.

## How it works

An MCP client connects to the local server, and the agent calls one of the exposed tools. `elicit_confirm` presents a yes-or-no style request such as whether to ship a change. `elicit_form` accepts a JSON Schema defined by the caller, allowing the agent to request several structured fields instead of relying on free-text replies. `elicit_doctor` checks the connected host's elicitation capability so an agent can identify compatibility problems before depending on user prompts.

Only an MCP server can initiate an elicitation request, which is why the local process exists even when the agent itself is running in another MCP client. Host support is not uniform, so the diagnostic result can help explain why an elicitation flow may not work in a particular environment.

## Setup and configuration

The elicitly MCP server runs directly from npm without requiring a repository checkout:

```sh
npx -y elicitly
```

Register that command with an MCP host that supports stdio. For Claude Code, the repository documents:

```sh
claude mcp add elicitly -- npx -y elicitly
```

Claude Desktop users can instead download the `elicitly.mcpb` file from the latest GitHub release and open it as a desktop extension. The package also supports embedding: install `@elicitly/tools` and register its elicitation tools on an application-owned `McpServer`. The provided material does not specify environment variables or additional runtime configuration.

## Tools and capabilities

The elicitly MCP server provides these documented tools:

- `elicit_confirm`: asks the user to confirm a requested action.
- `elicit_form`: collects user responses against a caller-defined JSON Schema.
- `elicit_doctor`: tests whether the current MCP host supports elicitation.

The npm package is intended for running the complete local server, while `@elicitly/tools` is intended for integrating the tools into another MCP server.

## Limitations and notes

The local Free Edition depends on elicitation support in the MCP host. A host may connect successfully while still lacking the capability needed for prompts or forms; `elicit_doctor` exists to make that condition visible. The README does not document support for arbitrary remote transports, authentication settings, persistence, or an audit trail for the local package. Durable approvals, review pages, audit trails, and organization dashboards belong to the separate hosted Pro Edition.

The elicitly MCP server is licensed under Apache-2.0. The repository also includes development commands for installation, building, testing, coverage, type checking, and formatting checks.

_Full upstream README: https://allmcps.com/mcp/elicitly/readme_

