AI Reserve

AI Reserve for GitHub Copilot

Two things you can do with GitHub Copilot and AI Reserve, and most teams do both:

1. Run Copilot on AI Reserve models. Our models in Copilot's own model picker, for Chat, agent mode and edits. Any AI Reserve key works and is what pays: whatever the key belongs to (a developer, a team, the organization) is debited, with its spend caps and model rules applied. Three ways in, differing only in who sets them up: the AI Reserve VS Code extension (each developer, one minute), organization-wide custom models (the GitHub admin, once; needs GitHub Enterprise; covers github.com chat too), or the manual Custom Endpoint / Copilot CLI variables. Step by step in ADMIN-INSTALL-GUIDE.md and on https://docs.aireserve.com/connect/github-copilot.
2. The AI Reserve assistant in Copilot Chat (the rest of this package). Tools inside the Copilot developers already use: ask any of our models by name ("using fable 5.1 ping me"), see prices, check usage and budget, manage API keys. Sign in with the portal account, nothing to paste.

GitHub retired App-based Copilot Extensions (the @vendor apps) on 10 November 2025 in favor of the Model Context Protocol. The assistant is therefore not an app bundle to upload: it is the install surface for the AI Reserve MCP server, which Copilot reads directly. The server is live at https://api.aireserve.com/mcp and signs users in over OAuth (dynamic client registration), so there is no key to distribute for IDE use.

Where it works

Our models in the picker (1):

The assistant (2):

Copilot Business and Enterprise: a GitHub admin must switch on the MCP servers in Copilot policy once (it is off by default). Everything else in the assistant table works on Copilot Pro and Free without admin involvement.

What is in the package

No credential lives in this package. IDE users sign in; the cloud agent's token is minted in the portal by whoever configures that repository and stored as a GitHub Agents secret.

Install

VS Code

Click Install when VS Code asks, then Allow when it asks to sign in to AI Reserve; a browser tab signs you in with your portal account. Or copy vscode/mcp.json to .vscode/mcp.json in the workspace (commit it and every clone gets the assistant) or to your user profile's mcp.json. VS Code 1.99 or newer; verify with MCP: List Servers.

JetBrains, Visual Studio, Eclipse, Xcode

Open Copilot Chat in Agent mode, click the tools icon (_Configure your MCP server_) → Add MCP Tools, paste the contents of vscode/mcp.json, and click Allow on the sign-in prompt. In Visual Studio, click Auth in the CodeLens above the server if the prompt does not appear.

Copilot CLI and the GitHub Copilot app

copilot mcp add --transport http aireserve https://api.aireserve.com/mcp

The first use opens the sign-in in your browser (Copilot CLI 1.0.11 or newer). The GitHub Copilot desktop app picks the server up automatically; to add it from the app instead, open Customize → MCP → Add custom server, choose HTTP, paste https://api.aireserve.com/mcp, leave the headers empty and sign in when prompted. copilot-cli/mcp-config.json is the equivalent file form.

github.com: Copilot cloud agent and code review

The cloud agent runs without a browser, so it carries a token:

1. In the AI Reserve portal, open API Keys → Management Keys and mint an MCP access token with the Messages scope (https://portal.aireserve.com/keys?type=mcp).
2. In the repository, add it as an Agents secret named COPILOT_MCP_AIRESERVE_TOKEN (Settings → Environments → copilot). GitHub exposes only COPILOT_MCP_* names to MCP servers.
3. Paste cloud-agent/mcp.json under Settings → Copilot → MCP servers and save. Tools are allowlisted explicitly because the agent calls them without asking first; send-message spends from the token owner's budget.

Try it

In Copilot Chat, switch to Agent mode and type:

using fable 5.1 ping me

Copilot resolves the model with list-models, sends the message with send-message, and answers with the reply and the exact model that produced it: Model: claude-fable-5-1. The same loop is one command, /mcp.aireserve.ask-model, and #send-message pins the tool. Other asks that work out of the box: "which models can I use and what do they cost", "what did my team spend this week", "how much budget do I have left", "list my API keys". Each answer is scoped to what the signed-in person can already see in the portal; anything that spends money or changes the account asks for confirmation. Full surface: MCP tool reference.

Models in the Copilot picker (the other half)

Separately from the assistant, two ways to put our models in Copilot's own model picker:

• Each developer, one minute: install the AI Reserve VS Code extension (download the .vsix from https://portal.aireserve.com/connect/copilot until the Marketplace listing is live), open the picker → Manage Models → AI Reserve, paste any key from the portal's API Keys page, and say yes when it offers to make AI Reserve the default for new chats. New catalog models show up on their own; the list refreshes every five minutes.
• The GitHub admin, once (GitHub Enterprise): add AI Reserve as an OpenAI-compatible custom model provider so our models appear in every developer's picker, including Copilot Chat on github.com, with nothing to install. The model list there is the admin's: when AI Reserve adds a model, add its id to that list.

Both are step by step in ADMIN-INSTALL-GUIDE.md and on https://docs.aireserve.com/connect/github-copilot.

Publishing to the registries

registry/server.json is the manifest both registries consume, named io.github.Audacity-Investments/aireserve so the repository's own GitHub Actions OIDC identity authorizes the publish (no secret, no PAT). The MCP Registry publish workflow (.github/workflows/mcp-registry-publish.yml) validates the manifest against the live registry on every run and, in publish mode, publishes it to the official MCP Registry (registry.modelcontextprotocol.io), which VS Code's MCP gallery and the GitHub Copilot app's catalog read from. Version 1.0.0 is published and active; bump version to publish again.

Listing on GitHub's MCP Registry page (github.com/mcp) is a manual curation step on GitHub's side after that publish; the workflow's job summary prints the onboarding request to post in the github/github-mcp-server discussions (or to send to partnerships@github.com). Once onboarded, later versions sync automatically. Until then, a customer that restricts MCP to an approved registry adds t