get_system_info
Return the system and environment details of the machine the server runs on, as JSON.
LABSCO SUMMARY
The agent asks at the point where it would otherwise pick an interpretation.
What it takes off you is the correction afterwards: the model hands you its summary while the work is still open, and your answer arrives as a tool result. Worth checking your client first — MCP now has native elicitation and interactive tool UIs, and if those cover the checkpoint you want, you do not need this.
What it is
A two-call server that puts a person in the loop. The feedback call posts the agent's own summary of what it has done against a project directory, then waits for your reply or for the timeout you set. The reply surface is either a browser page or a native desktop application, so it works locally, over SSH and inside WSL.
What you get
- A feedback prompt the agent raises mid-task, carrying its summary of the work and the project directory it applies to (
interactive_feedback) - A timeout on that prompt, so an unanswered request returns instead of holding the run open
- Text, saved prompts and uploaded images as the reply — images come back as standard MCP image content
- Two interfaces with the same functionality: a Tauri desktop application on Windows, macOS and Linux, or a browser page that needs no GUI on the machine — which is what makes it work over SSH and in WSL
- Session history and statistics recorded as the conversation goes
- The system and environment details of the machine the server runs on, returned as JSON (
get_system_info)
Requirements
uvx on your PATH, and someone at the keyboard — the feedback call waits for an answer, and what it returns is whatever you type. Run uvx mcp-feedback-enhanced@latest and stay on v2.6.1 or newer: that release removed the command-execution feature behind an unauthenticated WebSocket, and started validating Origin before accepting a connection so a web page you visit cannot reach the local /ws. The Web UI and its WebSocket endpoint still have no authentication of their own, so leave MCP_WEB_HOST at 127.0.0.1 and use SSH port forwarding for remote work rather than binding 0.0.0.0. MCP_WEB_PORT defaults to 8765, MCP_DESKTOP_MODE switches to the desktop application, and MCP_LANGUAGE forces English, Traditional or Simplified Chinese.
Setup effort
One command — uvx mcp-feedback-enhanced@latest
USE CASES
01
Signing off before the agent moves on
The agent posts what it thinks it has done and waits. Approving takes a sentence; the alternative is reading a diff after four more steps were built on top of a wrong assumption.
02
Redirecting a run that went the wrong way
Type the correction into the prompt and it comes back as the tool result, so the agent continues from your answer rather than restarting from a new instruction.
03
Showing the agent a screenshot instead of describing it
Upload the image in the reply. It returns as standard MCP image content, so the model looks at the broken screen rather than at your account of it.
04
Answering an agent that is running on a remote box
The browser interface needs no GUI on the host, which is what makes it work over SSH and in WSL. Forward the port over SSH rather than binding it to every interface — the page and its socket have no authentication.
05
Reporting the machine the agent is running on
Ask for the system and environment details as JSON when a bug looks environmental. It settles the platform question without a round trip through a person.
2 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
interactive_feedback
Interactive feedback collection tool for LLM agents.
USAGE RULES:
- During any process, task, or conversation, whether asking, replying, or completing phased tasks, you must call this tool to ask for feedback.
- Unless receiving termination instructions, all steps must repeatedly call this tool.
- Whenever user feedback is received, if the feedback content is not empty, you must call this tool again and adjust behavior based on the feedback content.
BEFORE YOU INSTALL
What to know first
Tools you get
2 tools when we connected. It is a human-in-the-loop gate, so two is the whole set. submit_feedback and wait_for_feedback in the API reference are a WebSocket message type and an internal coroutine, not callable tools.
Requires
uvx on PATH
CONTEXT COST
Light Standing cost
Rated by the size of the 2 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Connect your client
Server setup
TERMINAL
uvx mcp-feedback-enhanced@latestMCP Clients
Cursor, Cline, Windsurf, Other MCP clients
JSON
{
"mcpServers": {
"mcp-feedback-enhanced": {
"command": "uvx",
"args": [
"mcp-feedback-enhanced@latest"
],
"timeout": 600,
"autoApprove": [
"interactive_feedback"
]
}
}
}Any MCP client — web UI
ANY MCP CLIENT — WEB UI · JSON
{
"mcpServers": {
"mcp-feedback-enhanced": {
"command": "uvx",
"args": [
"mcp-feedback-enhanced@latest"
],
"timeout": 600,
"env": {
"MCP_DEBUG": "false",
"MCP_WEB_HOST": "127.0.0.1",
"MCP_WEB_PORT": "8765",
"MCP_LANGUAGE": "en"
},
"autoApprove": [
"interactive_feedback"
]
}
}
}Any MCP client — desktop app
ANY MCP CLIENT — DESKTOP APP · JSON
{
"mcpServers": {
"mcp-feedback-enhanced": {
"command": "uvx",
"args": [
"mcp-feedback-enhanced@latest"
],
"timeout": 600,
"env": {
"MCP_DESKTOP_MODE": "true",
"MCP_WEB_HOST": "127.0.0.1",
"MCP_WEB_PORT": "8765",
"MCP_DEBUG": "false"
},
"autoApprove": [
"interactive_feedback"
]
}
}
}Install check
INSTALL CHECK · TERMINAL
uvx mcp-feedback-enhanced@latest test --web
