Stable Baseline

Client recipes

Gemini with Stable Baseline.

Google Gemini connects to remote MCP servers through the Gemini CLI (or a Gemini Extension). Point it at Stable Baseline and Gemini can read and write your whole workspace.

Two paths, same result

You can connect Gemini to Stable Baseline two ways:
  • Gemini CLI — fastest path; paste a JSON block into the settings file and you're done.
  • Gemini Extension — packages the same config as a shareable bundle your whole team can install.
Start with the CLI; graduate to an Extension when you want a repo-committable setup.

Before you begin

Step 1 · Export your API key

Keep the key out of the settings file by putting it in your shell profile (~/.bashrc, ~/.zshrc, etc.):

Shell profilebash
export STABLE_BASELINE_API_KEY="sta_your_api_key_here"

Step 2 · Gemini CLI — add the MCP server

Gemini CLI reads MCP server definitions from ~/.gemini/settings.json. Open (or create) that file and drop in:

~/.gemini/settings.jsonjson
{
  "mcpServers": {
    "stable-baseline": {
      "httpUrl": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
      "headers": {
        "Authorization": "Bearer $STABLE_BASELINE_API_KEY"
      }
    }
  }
}

Save, start Gemini CLI with gemini, and run the built-in /mcp command. Stable Baseline should appear in the list with its tools loaded.

Per-project overrides

Gemini CLI also reads .gemini/settings.json from the current directory, so you can check a project-scoped config into your repo and have teammates get the server automatically.

Step 3 (optional) · Wrap it as an Extension

If you want the same setup to ship across an org (or install one-click for teammates), create a Gemini Extension. The manifest format wraps the same MCP block you already wrote:

gemini-extension.jsonjson
{
  "name": "stable-baseline",
  "version": "1.0.0",
  "mcpServers": {
    "stable-baseline": {
      "httpUrl": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
      "headers": {
        "Authorization": "Bearer $STABLE_BASELINE_API_KEY"
      }
    }
  }
}

Publish the extension (private or public) and your team runs gemini extensions install stable-baseline. No secrets in the manifest — the bearer token is still sourced from the env var.

Step 4 · Verify the connection

Inside Gemini CLI, run:

Gemini CLIprompt
/mcp list

You should see stable-baseline with a healthy status. Then run a real probe:

Gemini CLIprompt
Using the stable-baseline MCP server, list my workspaces and projects.

Step 5 · Try a real task

“Find all open improvements in my Platform project and summarise the blockers for each.”

“Read the PRD for Multi-region GA and draft a launch checklist. Save it as a new document.”

“Create a BPMN diagram for our incident response process with lanes for on-call, SRE, and comms.”

Gemini app shortcut

The consumer Gemini web app (gemini.google.com) currently surfaces extensions under Apps / Extensions → Manage. If your organisation has deployed the Stable Baseline extension, it will appear there and you can enable it per-chat the same way you would for any Google app.

Consumer web app: still rolling out

Full MCP support inside the Gemini web chat is rolling out gradually via Extensions. If the Extensions toggle is missing in your account, use Gemini CLI — it has full MCP support today.

What works, what doesn't

  • Gemini CLI — full 60+ MCP tools, persistent /mcp list, per-project configs. Recommended.
  • Gemini Code Assist (IDE) — supports MCP via the same settings.json format. Works in VS Code and JetBrains.
  • Gemini web app — Extensions path is stable for some accounts, rolling out for others.
  • Gemini API directly — if you're calling the Gemini API from your own code, wire Stable Baseline in as a mcp-remote stdio bridge or call our HTTP endpoints from your own tool definitions.

ChatGPT recipe

OpenAI's chat over custom MCP.

Claude Code recipe

Anthropic's agentic CLI.

Permissions

Scope per agent, not per user.

Tools reference

Every MCP tool available.