MCP for AI agents

Using VS Code? Choose the agent first

Install and sign in to the extension you intend to use. Configure SetGet for that extension, then run the first-task prompt in its chat panel. Choosing a Claude model in Copilot does not switch you to the Claude Code extension.

With WSL, SSH or a dev container, confirm where the agent runs. Its loopback address, home directory and environment may differ from your desktop’s. Use an endpoint reachable from that environment and configure credentials there.

Claude Code in VS Code or the terminal

In the Claude Code extension, open your project and type /mcp in its chat panel to manage servers. You can also run the commands below in VS Code’s integrated terminal with the Claude CLI installed. Start a new conversation after changing configuration, then use /mcp to check that SetGet is connected.

Source: Claude Code’s VS Code guide.

Run these commands in your project terminal, then use /mcp inside Claude Code to inspect the connection and available tools.

claude mcp add --transport http setget http://127.0.0.1:8080/mcp
claude mcp get setget

For an existing protected key, use this entry in your project’s .mcp.json instead. Merge it with any existing servers. Set SETGET_TOKEN privately in the environment that launches the client; leave the variable reference in the file.

{
  "mcpServers": {
    "setget": {
      "type": "http",
      "url": "http://127.0.0.1:8080/mcp",
      "headers": {"X-SetGet-Token": "${SETGET_TOKEN}"}
    }
  }
}

Source: Claude Code’s official MCP instructions.

Codex in VS Code

Open the Codex extension’s gear menu → MCP servers → Add server. Name it setget, choose Streamable HTTP, and enter http://127.0.0.1:8080/mcp (or your selected deployment). Save, then choose Restart extension. Check that SetGet is enabled and try the first-task prompt in Codex.

Alternatively, merge this into ~/.codex/config.toml on the Codex host. CLI and IDE extension share this configuration; trusted projects can use .codex/config.toml. Edit an existing SetGet table rather than adding it twice.

[mcp_servers.setget]
url = "http://127.0.0.1:8080/mcp"

For an existing protected key, add this line inside that table and provide SETGET_TOKEN in the Codex host environment before restarting:

env_http_headers = { "X-SetGet-Token" = "SETGET_TOKEN" }

If the CLI is installed, codex mcp list lists configured servers; /mcp in the CLI shows active connections. SetGet does not require OAuth login.

Source: Official OpenAI MCP documentation.

GitHub Copilot in VS Code

In Chat, choose your intended session target. Run Chat: Open Customizations from the Command Palette, open MCP Servers, and choose Add Server → HTTP. Enter your selected endpoint and name it setget. For personal Copilot sessions, choose Global → Copilot Global. Enable the tools under Chat: Open Customizations → Tools, then try the first-task prompt.

For a Local harness session, merge this into .vscode/mcp.json. This VS Code format uses servers; Copilot’s portable format uses mcpServers. Current VS Code labels Local configuration destinations deprecated in its guided flow, but retains them for compatibility.

{
  "servers": {
    "setget": {
      "type": "http",
      "url": "http://127.0.0.1:8080/mcp"
    }
  }
}

For protected keys in a Local session, use this complete configuration instead. It prompts for the capability without putting its value in the file:

{
  "inputs": [
    {
      "id": "setget-token",
      "type": "promptString",
      "description": "SetGet key capability",
      "password": true
    }
  ],
  "servers": {
    "setget": {
      "type": "http",
      "url": "http://127.0.0.1:8080/mcp",
      "headers": {"X-SetGet-Token": "${input:setget-token}"}
    }
  }
}

Interactive input configurations are not forwarded to Copilot harness sessions. For protected access there, use its documented environment-based header configuration; the prompt configuration above is Local-only. Source: VS Code setup and configuration reference.

Other VS Code extensions and editors

Open the extension’s MCP settings and look for a remote HTTP or Streamable HTTP server. Enter the selected /mcp endpoint; for protected keys, configure X-SetGet-Token through its supported secret mechanism. Enable the server and check that create_key and get_key appear before trying the first-task prompt.

Configuration files and secret syntax are extension-specific. A stdio-only client needs an additional bridge, which this guide does not configure. Other extensions remain untested. For Cursor, use the dedicated instructions below.

Cursor

Merge this into .cursor/mcp.json in your project. Open Cursor’s MCP settings, enable SetGet, and check that its tools appear.

{
  "mcpServers": {
    "setget": {"url": "http://127.0.0.1:8080/mcp"}
  }
}

For an existing protected key, add this property alongside url. Make the environment variable available to Cursor before launching it.

"headers": {"X-SetGet-Token": "${env:SETGET_TOKEN}"}

Source: Cursor’s official MCP instructions.

Python SDK

With Python 3.10 or newer and curl installed, create an isolated environment, download the demo, then run it against your selected server. On Windows, use the virtual environment’s Scripts/python.exe instead of bin/python.

python3 -m venv .venv-setget
.venv-setget/bin/python -m pip install 'mcp==2.3.0'
export SETGET_BASE=http://127.0.0.1:8080
curl --fail --show-error "$SETGET_BASE/static/examples/mcp_roundtrip.py" -o mcp_roundtrip.py
.venv-setget/bin/python mcp_roundtrip.py "$SETGET_BASE/mcp"

Expected output: Created and read a public one-hour key successfully. The script then deletes that key; a cleanup failure exits with an error. Download source.