Skip to content
| Marketplace
Sign in
Visual Studio Code>Machine Learning>LunaRoute for VS CodeNew to Visual Studio Code? Get it now.
LunaRoute for VS Code

LunaRoute for VS Code

Preview

LunaRoute

|
1 install
| (0) | Free
LunaRoute models in native VS Code Chat and Agent - browser sign-in, no API keys to copy.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

LunaRoute for VS Code

Your LunaRoute models, inside native VS Code Chat and Agent — with a browser sign-in instead of an API key to copy around.

Pick a LunaRoute model in the same model picker you already use, work in Chat or Agent mode, and let VS Code keep owning tools, approvals and workspace access. One sign-in connects the organization your plan belongs to.

Preview. This extension is in active preview. Verified today: sign-in and native Chat on desktop VS Code 1.141 (Linux), model discovery, streaming, tool calls, and LunaRoute's hosted tools through MCP. Still being validated: Windows and macOS extension hosts, remote workspaces (SSH/WSL/Dev Containers), long Agent sessions with tool history, and Marketplace install paths. Details and current evidence live in the repository's smoke checklist.

What you get

Native, not a side panel LunaRoute models appear in VS Code's own Chat model picker. No separate chat UI to learn, no third-party extension reconfiguration.
Agent mode with real tools Tool calls, approvals and execution stay with VS Code. Switch models or modes without changing how you work.
Hosted LunaRoute tools Web, image and document tools from LunaRoute's hosted MCP server are offered to Agent mode — enable or disable them in MCP SERVERS - INSTALLED (or MCP: List Servers).
Sign in, don't paste keys Browser authorization writes the credential straight into VS Code's SecretStorage. Nothing lands in a file, and Sign Out removes it locally.
Your plan, your policies Models and tools follow your organization's entitlement and policy — including per-plan model allowlists and per-tool MCP switches.
Reasoning models behave Reasoning is preserved for follow-up turns as opaque replay data and never shown as if it were the answer.
Honest failure modes No silent model switching, no automatic retry loops, no workspace-configurable endpoints. If something can't be represented, you get told.

Quick start

  1. Install this extension in desktop VS Code 1.141 or later.
  2. Run LunaRoute: Sign In from the Command Palette, then approve the request in the browser. Wait for the “Connected to LunaRoute” notification — that, not the browser redirect, means the credential is saved.
  3. Open Chat, choose a LunaRoute model, and start typing. Switch to Agent mode when you want tools.

Commands you'll actually use:

  • LunaRoute: Sign In — connect an organization (repeat to switch).
  • LunaRoute: Manage — see the connection, refresh models, open the dashboard, sign out.
  • LunaRoute: Refresh Models — reload the model list from the gateway.
  • LunaRoute: Sign Out — remove the local credential. The key stays active server-side until you revoke it in the dashboard.
  • LunaRoute: Capture Chat Diagnostics (Preview) — one-shot, local capture of a single request/response pair when support asks for it.

Requirements

  • Desktop VS Code 1.141+, extension installed in the local (UI) host. Browser-only VS Code (vscode.dev) is not supported; on Remote SSH / WSL / Dev Containers install it locally, not in the container.
  • A browser on the same computer as VS Code — sign-in briefly listens on 127.0.0.1 for the callback.
  • A LunaRoute account with the organization that should pay for the usage.
  • Network access to app.lunaroute.com, api.lunaroute.com, gw.lunaroute.com and mcp.lunaroute.com.

Credentials, privacy and limits

  • The dedicated lunaroute-vscode key lives only in VS Code SecretStorage. There is no plaintext fallback, no workspace-configurable endpoint, and this extension never writes the key to disk.
  • Requests resend the full history with store: false. The extension keeps no conversation store and never logs credentials, callback URLs, prompts or conversation content.
  • Sign Out removes the local connection only — revoke unused keys in the LunaRoute dashboard, including any key left behind by a cancelled or failed sign-in.
  • Token counts shown to the model picker are approximate (bytes plus framing and image budgets), not billing counts. Images count as 16,384 tokens each.
  • Optional diagnostics capture writes raw request/response files only to a folder you choose, and never includes request headers (so never your key).

Troubleshooting

Symptom Do this
Sign-in never returns Use the approval page's manual Return to VS Code link. Approval codes expire after 60 seconds — start a new sign-in if it lapsed.
“unknown device-auth client” or a gateway error during sign-in Update to the current preview build; older builds predate server-side support.
401 / revoked key Run Sign In again. The model list is invalidated, credentials are not silently deleted.
403 / policy Check organization and model permissions, and any desktop AI policy in your organization.
402 / billing Open Manage → dashboard and finish the connected organization's plan setup.
429 / rate limited Wait for the stated retry interval, then retry explicitly — nothing retries for you.
No models listed LunaRoute: Refresh Models; check connectivity to gw.lunaroute.com.
“unsupported or oversized input” Reduce the input or start a new chat — unsupported history fails explicitly rather than disappearing.
MCP tools missing MCP: List Servers → enable LunaRoute. If it reports unavailable, check its output before editing mcp.json by hand.
Something else Run LunaRoute: Capture Chat Diagnostics (Preview), reproduce once, and send the captured files to LunaRoute support.

Chat without GitHub sign-in

LunaRoute authentication and model access don't require a GitHub session. VS Code may still want utility models for background chat work — pick LunaRoute models in Settings (chat.utilityModel, chat.utilitySmallModel), or set chat.byokUtilityModelDefault to mainAgent. Those are VS Code settings; this extension never changes your preferences, and background requests can consume your allowance.

About LunaRoute

Fixed-price inference for supported open models — so another attempt doesn't cost another token. Learn more at www.lunaroute.com and in the docs.

License

MIT — Copyright (c) 2026 LunaRoute, Inc. License and bundled third-party notices ship with the extension (LICENSE, THIRD_PARTY_NOTICES.txt).

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
  • Your Privacy Choices
  • Consumer Health Privacy
© 2026 Microsoft