Skip to main content
Glama
alexVirtualWorld

Local Agent MCP

by alexVirtualWorld
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

Local Development Agent — ChatGPT MCP version

  • shell execution

  • workspace file read/write/list

  • screenshot capture, now returned to ChatGPT as MCP image content

  • mouse clicks and keyboard shortcuts

  • persistent Markdown memory files

  • skill promotion workflow

  • persistent context loader for identity / memory / skills / recent daily logs

Video Tutorial

A complete step-by-step tutorial is available on YouTube:

GPT Web Agent Setup Tutorial

▶ Watch the tutorial on YouTube

Related MCP server: Chat On Steroids

Install

Python 3.10+ is required by the current MCP Python SDK.

Open PowerShell 1：

  1. Enter the directory like this： cd E:\AI_agent\chatgptweb

  2. input install_mcp.bat 3.input start_mcp.bat

Open Web：

1.Download tunnel-client https://github.com/openai/tunnel-client/releases/tag/v0.0.14

2.Extract the files and copy them to the folder containing `agent.py

3.1Register for an OpenAI Platform API key

https://platform.openai.com/api-keys?utm_source=chatgpt.com

3.2Create new secret key

4.1Tunnel Settings： https://platform.openai.com/settings/organization/tunnels?utm_source=chatgpt.com

4.2Create a tunnel_id

Open PowerShell 2：

  1. $env:CONTROL_PLANE_API_KEY="Your OpenAI API key"

  2. $env:CONTROL_PLANE_TUNNEL_ID="Your tunnel_id"

  3. $env:MCP_SERVER_URL="url=http://127.0.0.1:8001/mcp,channel=main"

  4. .\tunnel-client-runtime.exe run

Open chatgpt web setting → Security and login → Developer mode

Plugins → New Plugin

Connection select Tunnel Switch to using tunnel ID. Authentication select No Auth I understand and want to continue Creat

Open ChatGPT Web

send a message： Use the Local Agent MCP to call list_workspace and tell me what is in the current workspace. Do not modify any files.

example and test： Use the Local Agent MCP to read AGENTS.md and IDENTITY.md, and load the agent context. Tell me the current rules and long-term memory for this agent. Do not modify any files.

Use the Local Agent MCP to call take_screenshot, capture my current screen, and tell me what you see.

Security notes

run_shell, write_file, mouse, keyboard, and memory-write tools can change the local machine. Only connect this MCP server to an OpenAI workspace/account you trust. The MCP annotations mark consequential tools accordingly, but annotations are metadata, not a security boundary.

The file read/write tools are constrained to workspace/. run_shell starts inside workspace/, but shell commands are intentionally powerful and can still access the broader machine unless you additionally sandbox the Python process at the OS/container level.

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • ChatGPT Local Coder

    Mieruko
    A
    license
    Not graded
    quality
    B
    maintenance
    Turns ChatGPT web into a local coding agent, enabling file edits, shell commands, Git operations, patches, and process management through 40+ MCP tools.
    MIT

  • Chat On Steroids

    DaveNguyen2608
    A
    license
    Not graded
    quality
    B
    maintenance
    Gives ChatGPT hands-on control of a local machine: reading, patching and executing code in approved project folders, running real interactive processes, managing plans and worker sub-chats, and driving the browser or desktop through user-granted permissions. Every capability is a switch inside an approved-folder sandbox, so nothing runs that the user did not explicitly enable.
    1
    MIT

  • LocalPilot

    a252937166
    F
    license
    Not graded
    quality
    C
    maintenance
    Enables ordinary ChatGPT web Chat to work directly on a local machine over a private MCP tunnel — reading and editing project files, running shell commands and tests, loading local skills, manipulating images, and driving Chrome — without spinning up a separate coding agent or additional model-generation API calls. It also supports multi-step task plans with persisted execution receipts and acceptance checks so results from real runs are returned to the conversation.
    -

  • Core

    shadorig
    A
    license
    Not graded
    quality
    A
    maintenance
    Lets ChatGPT work directly on a local project by reading and editing files, running shell commands and tests, keeping terminals open, and using the user's desktop, all paired with an approved workspace folder. It also splits independent jobs across persistent workers that retain context, while riding the user's existing ChatGPT plan rather than Codex quota.
    MIT