rishvan-mcp

A stdio MCP server that provides a human-in-the-loop tool (ask_rishvan) with a web UI for responding to LLM questions.

How it works

1. An LLM client calls ask_rishvan(question, app_name) via MCP stdio
2. The server creates a pending request in SQLite (scoped to the IDE), starts a web UI on port 56234, and opens the browser
3. The human sees the question in the React UI and types a response
4. The response is returned to the LLM as the tool result

Multiple IDEs can use the same server binary — each source passes --source <name> and gets isolated state.

Build

Requires Go 1.23+ and Node.js 18+.

make

This builds the frontend, embeds it into the Go binary, and produces ./rishvan-mcp.

Usage

rishvan-mcp --source <source-name>

The --source flag is required. It scopes all DB records and UI state to that source instance.

MCP Configuration

Add to your MCP client config (e.g. Claude Desktop, Windsurf):

{
  "mcpServers": {
    "rishvan": {
      "command": "/path/to/rishvan-mcp",
      "args": ["--source", "windsurf"]
    }
  }
}

Docker

docker pull tejzpr/rishvan-mcp:latest
docker run --rm -it tejzpr/rishvan-mcp --source windsurf

Images are published to Docker Hub on each GitHub release for linux/amd64 and linux/arm64.

Tool

ask_rishvan

Parameter	Type	Required	Description
question	string	yes	The question or prompt for the human
app_name	string	yes	Application/project context name

Returns: The human's text response.

Data

• Database: ~/.rishvan-mcp/app.db (SQLite via GORM)
• Web UI: http://localhost:56234