01 — UNDERSTANDRead
Understand the project first.
Let web AI inspect real directories, read files, and search code. No need to move the entire project into a chat or repeatedly paste in context.
Browse directories · Read files · Search code
VS CODE EXTENSION/ 0.1.34
Your web AI.
Local code.
Connected.
Connect MCP-capable web AI to your local VS Code project. Read code, edit files, run commands—and turn conversations into real development.
Open the extension page, then select Install.
Desktop VS Code 1.106.0+ · Valid user_key required to connect
LOCAL PROJECT, CONNECTED
Real interface · Click to enlarge
01 / CAPABILITIESRead the code.
Read the code.
Then do the work.
Describe a task in your MCP-capable web AI, then let EasyAI MCP connect it to your local project. Not another chat window, and not just a code snippet. Review actual changes right in VS Code.
02 — MAKE CHANGESEdit
Write changes to local files.
Go beyond suggested code. Use project MCP tools to edit files and apply changes, turning your discussion into work inside the current VS Code project.
Edit files · Apply changes · Work in your project
03 — EXECUTERun
Run commands. Follow the results.
Run development commands, tests, and builds. Read output, track jobs, and stop them when needed. Review tool calls in the activity panel.
Run commands · Test & build · Monitor jobs
AND MORE
Bring skills, external tools, and activity together.
- Skills · Reusable knowledge
- Search skills.sh or import existing Skills. Projects inherit globally installed skills, with per-project exceptions. AI reads skill instructions and text resources on demand; bundled scripts are not executed.
- Explore Skills ↗
- Third-party MCP · More tools
- Search the official Registry or import JSON to manage stdio and Streamable HTTP services. Configure globally, disable per project, and discover or call configured tools remotely.
- Explore tool management ↗
- Activity & diagnostics · Clearer context
- Review tool calls, states, and durations. Search or export activity details, then use diagnostics and logs to check the local environment and troubleshoot connections.
- Explore diagnostics ↗
02 / GET CONNECTEDConnect your project.
Connect your project.
Let web AI get to work.
Connect in the extension, then add your current project to web AI.
Your key and project link never need to be entered on this website.
Your browser may ask permission to open VS Code. Allow it, then confirm installation on the extension page.
Not opening? Visit the Marketplace ↗, or copy the command below.
code --install-extension mcp-oai9-com-EasyAI.easyai-mcpInstall the extension and reload
Use the install link, allow your browser to open VS Code, and select Install on the EasyAI MCP extension page. Reload the window if prompted.
Open and trust your local project
Open the project folder and confirm Workspace Trust so the extension can establish the current project's connection.
Open settings and enter your key
Enter a valid user_key in EasyAI settings. At this point, the extension only previews account information; it does not save the key or start the Runner.
Save and connect automatically
Select Save and auto-connect. The key is stored in VS Code SecretStorage, then the extension pairs and connects the project.
Copy the current project's MCP link
Wait for the project to come online, then copy the MCP link or the link and prompt. The link includes access credentials—keep it private.
Connect web AI and start developing
Add the project connection to MCP-capable web AI and ask it to read code, edit files, or run tests. Setup varies by AI client. Keep the VS Code window that owns the Runner online.
03 / IN PRACTICE
See how it works.
Start with the connection flow, then see real project work.
From the interface to practical development.
Connect your first project.
Watch the setup tutorial for MCP configuration, extension connection, and the basic workflow.
Put web AI to work on real code.
Watch the Bilibili walkthrough to see AI understand a project and help complete development tasks.
04 / GOOD TO KNOWBefore you connect,
Before you connect,
a few useful answers.
Know the connection's scope and limits.
They are part of working together.
Does clicking Install automatically install the extension?
No silent installation takes place. The button uses the VS Code protocol to open the EasyAI MCP extension page. Your browser may first ask to open VS Code; you still select Install in the editor and confirm publisher trust if prompted. If nothing opens, use the Marketplace alternative or the install command in Get started. You can read this site on a phone, but installation and local development require desktop VS Code.
Can any web AI operate on my local project?
The AI must support MCP connections and the relevant tool calls, and you must configure the current project's connection first. Once connected, it can use project tools to read, search, and edit files or run development commands. Arbitrary chat websites do not automatically gain local access. This is not remote desktop software.
Which systems are supported? Is installation enough?
Version 0.1.34 supports Windows x64, macOS Intel x64, and Apple Silicon arm64, with VS Code 1.106.0 or later. Connecting requires a valid EasyAI user_key. Installing the extension does not grant free or perpetual service. Native macOS components are currently ad-hoc signed and have not been notarized by Apple.
Can AI keep working after VS Code closes?
No. The VS Code window responsible for the project's connection must remain online. Reopening with the same machine, profile, project path, and account usually reuses the connection. Moving the project or changing devices or profiles creates a new identity.
Can one account connect multiple projects?
Yes. Each project has its own Workspace MCP connection. If the same project is open in several windows, only one window holds the Runner lease to avoid duplicate execution. This is not a filesystem sandbox.
How do Skills differ from third-party MCP services?
Skills provide instructions and text resources that AI can read on demand; bundled scripts are not executed. Third-party MCP services provide discoverable, callable tools. stdio and Streamable HTTP are supported; OAuth and legacy SSE are not. Third-party programs run with the current OS user's permissions, not in a sandbox. Remote third-party MCP calls do not receive a confirmation dialog for each invocation, so only configure trusted services.
How should I protect my project link and key?
Your user_key is an account credential, and your MCP capability URL contains project access credentials. Do not post either publicly, include them in public screenshots, or commit them to Git. If a link leaks, delete or recreate the corresponding Workspace Binding and update the connection in your AI client.
Let's make your next project work.
Need help with installation, connections, or multiple projects? Join the conversation.