Workspace .vscode/mcp.json uses servers. Copilot CLI uses mcpServers.
Jev in VS Code
Use Jev in VS Code by registering an MCP server and opening it in Copilot Agent mode.
1. Create the workspace file .vscode/mcp.json with a top-level servers entry.
For system1-mcp the command is uvx and the args are system1-mcp.
Or run MCP: Add Server from the Command Palette, choose Command (stdio), and save to that workspace file.
The same flow is Add Server under Chat: Open Customizations, then MCP Servers.
2. Save the file. Click Start at the top of the server list. When VS Code asks, confirm that you trust the server.
After you trust the workspace, servers in .vscode/mcp.json can start without a separate MCP trust prompt.
3. Open Copilot Chat and select Agent from the agent dropdown.
4. Open the tools list and confirm the Jev server's tools are there.
GitHub's VS Code steps call this control the tools icon. The VS Code MCP quickstart calls it Configure Tools.
VS Code also documents the workspace file .mcp.json. Its top-level key is mcpServers.
Copilot harness sessions read that file directly. They do not read .vscode/mcp.json directly.
They can receive a forwarded copy, except servers that use interactive input variables such as ${input:...}.
{
"servers": {
"system1": {
"command": "uvx",
"args": ["system1-mcp"],
"env": {
"TYPESAFE_API_KEY": "${input:typesafe-api-key}"
}
}
},
"inputs": [
{
"type": "promptString",
"id": "typesafe-api-key",
"description": "TypeSafe API key for Jev (never commit literals)"
}
]
}Agent mode for MCP tools
GitHub's VS Code walkthrough uses Agent mode for MCP. Open Copilot Chat from the chat icon, select Agent from the agent dropdown, make sure the server is running, then use the tools icon to see servers and tools. Select Agent when you want Copilot to call Jev MCP tools.
Which file uses servers
The top-level key depends on the file.
Workspace .vscode/mcp.json and the VS Code user profile file mcp.json use servers.
The workspace file .mcp.json and the Copilot user file use mcpServers.
A Cursor or Claude Desktop file that starts with mcpServers does not match .vscode/mcp.json.
Copilot CLI also skips .vscode/mcp.json because that file uses servers.
For a new shared server, VS Code prefers .mcp.json or Copilot Global.
The Add Server flow lists .vscode/mcp.json and the user-profile mcp.json as deprecated compatibility destinations.
The file .vscode/mcp.json is still what GitHub's VS Code steps use, and it is where VS Code input variables work.
Install system1-mcp for Copilot
system1-mcp documents four tools: fast_guard, fast_judge, fast_verify, and fast_score.
Its installer autodetects Claude Desktop, Cursor, Google Antigravity, Windsurf, Roo Code, Cline, and Zed.
It does not name Visual Studio Code, so uvx system1-mcp install is not the VS Code setup.
Add the server in .vscode/mcp.json yourself, then run uvx system1-mcp doctor.
If the key is already in ~/.system1/config.json, the README says the env block is optional.
Do not commit a live key. The sample above uses a VS Code input prompt.
Copilot harness sessions skip servers that use those input variables.
export TYPESAFE_API_KEY="tsk_..."
uvx system1-mcp doctortypesafe-mcp alternative
itsmostafa/typesafe-mcp, the system-one-connector README, exposes one evaluate tool for POST /v1/systemone shaped calls.
Install with the system-one-connector install script, then run evaluate setup mcp with TYPESAFE_API_KEY set.
That command registers Claude Code, Claude Desktop, Codex, and Hermes. It does not register Visual Studio Code.
For VS Code, the manual config runs the evaluate binary with args mcp.
The installer places that binary in ~/.local/bin unless EVALUATE_INSTALL_DIR is set.
Put the command under servers in .vscode/mcp.json.
The project example uses mcpServers and a literal env value. Change the key to servers for this file.
Use a VS Code input if you do not want the key in git. Pair it with typesafe-ai/skills when you want notes on when to call evaluate.
{
"servers": {
"evaluate": {
"command": "/absolute/path/to/evaluate",
"args": ["mcp"],
"env": {
"TYPESAFE_API_KEY": "${input:typesafe-api-key}"
}
}
}
}Copilot CLI and cloud coding agent
Copilot CLI user servers live in ~/.copilot/mcp-config.json under mcpServers.
When COPILOT_HOME is set, the user file is $COPILOT_HOME/mcp-config.json.
Otherwise the user file is ~/.copilot/mcp-config.json.
copilot mcp add writes ~/.copilot/mcp-config.json.
Project files are .mcp.json and .github/mcp.json.
They use mcpServers, or a bare object whose keys are server names.
The CLI does not read .vscode/mcp.json.
Copilot cloud agent does not use the VS Code file either.
A repository admin pastes an mcpServers JSON block on GitHub under Settings, then Copilot, then MCP servers.
Secrets and variables named in that block must use the COPILOT_MCP_ prefix.
Cloud agent and Copilot code review support MCP tools only.
They do not support MCP resources or prompts, and they do not support remote MCP servers that use OAuth.
A local command has to exist on the GitHub Actions runner.
The docs name uv and pipx as examples that are not installed by default and may need a copilot-setup-steps workflow.
Visual Studio and enterprise
Visual Studio 17.14 or later can use MCP with Copilot.
This is Visual Studio, not Visual Studio Code.
Open GitHub Copilot Chat, select Agent, click the tools icon, then the plus icon, fill in the server, and save.
The Visual Studio mcp.json examples use a top-level servers key for HTTP servers and for local stdio servers.
VS Code says organizations centrally manage MCP server access with Copilot enterprise-managed settings.
GitHub has a separate MCP servers in Copilot policy for Copilot Business and Copilot Enterprise.
That policy is disabled by default.
It does not govern Copilot Free, Copilot Pro, Copilot Pro+, or Copilot Max.
Environment keys
Create a TypeSafe API key in the console linked under Primary sources.
You can also route through Vercel AI Gateway or OpenRouter where the integration documents it.
Export TYPESAFE_API_KEY for VS Code Copilot Agent and MCP child processes.
Some MCP clients accept AI_GATEWAY_API_KEY for gateway-hosted Jev. Follow that integration README.
Never commit keys into repos or paste them into chat logs.
export TYPESAFE_API_KEY="tsk_..."
# Optional pin:
export TYPESAFE_MODEL="jev-latest"Open models beside hosted Jev
Copilot Agent mode calls hosted Jev through MCP today. When you compare self-hosted System One compatible servers for other checks in your workflow, our dated pages cover migration and benchmarks. Start from the ranked hub when you want the full list.
See Jev alternatives compared, Liquid d1 vs Jev and Kev vs Jev.
Pitfalls
Do not commit live API keys.
In .vscode/mcp.json, use inputs so Start prompts for the key.
Copilot harness sessions do not receive servers that use input variables.
Copilot CLI does not read .vscode/mcp.json.
After edits, click Start or run MCP: List Servers.
A server that never started has no tools in Agent mode.
The system1-mcp README says registration alone does not force a call to fast_guard.
It suggests an AGENTS.md line that calls fast_guard before shell commands.
Related directory listings
Related agent guides
Cline and Roo Code + Jev
MCP reflex tools in the VS Code sidebar agent. system1-mcp installer or typed evaluate servers.
Cursor, Grok, and agent skills + Jev
MCP in Cursor settings, TypeSafe skills in the agent, no imaginary Grok-only APIs.
Devin Desktop + Jev
Windsurf became Devin Desktop. MCP reflex beside Devin Local, Claude, and Codex agents.
Codex and OpenCode + Jev
jev-judge on the CLI, evaluate setup, and routers that treat Jev as traffic control.
Jev with Hermes Agent
Smart approvals, skill routing, and MCP without asking Hermes to fake a classifier.
FAQ
- How do I use Jev in VS Code?
- Add a Jev MCP server in workspace .vscode/mcp.json under servers, or run MCP: Add Server and choose Command (stdio). Click Start, trust the server or the workspace, open Copilot Chat, select Agent, and confirm the tools in the tools list.
- What is Jev in VS Code?
- Jev in VS Code is an MCP server Copilot Agent mode can call, usually system1-mcp or the evaluate server. It is not a built-in VS Code extension listed on this site.
- What does VS Code Jev mean?
- VS Code Jev means Jev used inside Visual Studio Code through an MCP server and Copilot Agent mode. Follow the Jev in VS Code steps on this page.
- Why did my Cursor MCP config not work?
- Cursor configs use mcpServers. VS Code's .vscode/mcp.json file requires servers. Rename the key for that file, or use workspace .mcp.json when you want to keep mcpServers for Copilot harness sessions.
- Does Copilot Agent call Jev automatically?
- Only after the MCP server is configured, started, and trusted, and only when Agent mode can see its tools. VS Code may still ask you to confirm a tool call. The system1-mcp README says to add an AGENTS.md line that calls fast_guard before shell commands, because registration alone does not force the call. Copilot cloud agent can use enabled tools without asking once the repository MCP configuration is saved.
- Can I use jev-judge-mcp?
- Yes. The Claude and Codex guide documents stdio npx -y jev-judge-mcp. In VS Code .vscode/mcp.json, put that command under servers.
- Cloud agent MCP support?
- Repository admins add an mcpServers JSON block under Settings, Copilot, MCP servers. That is not .vscode/mcp.json. Cloud agent supports MCP tools only, not resources or prompts, and it does not support remote OAuth servers.
- Where is official Copilot MCP documentation?
- See GitHub Docs: Extending Copilot Chat with MCP, and the VS Code page Add and manage MCP servers. Visual Studio setup on the GitHub page starts at version 17.14.