Integrations: MCP server and GitHub Action

MCP server for coding agents

Coding agents know SDKs and models only up to their training date. Connect the Versyka MCP server and they can check what is current right now: the latest version of an SDK, whether a model is retired, and the step-by-step guide for a breaking upgrade, with the vendor sources.

Server address: https://www.versyka.com/api/mcp. It uses the Streamable HTTP transport, needs no sign-in and is read-only.

Connect your agent

Pick your assistant. Each takes about a minute, and nothing needs an account or a key.

Claude Code

Open a terminal in your project. (No Claude Code yet? Install it with npm install -g @anthropic-ai/claude-code.)

Add the server for this project:

claude mcp add --transport http versyka https://www.versyka.com/api/mcp

Or for every project on this computer:

claude mcp add --transport http --scope user versyka https://www.versyka.com/api/mcp

Start claude and type /mcp. versyka should show as connected, with 7 tools.
Ask a question (see Using it). The first time Claude Code calls a Versyka tool, it asks your permission; choose to always allow it for this server.

Cursor

Open Cursor Settings → MCP (Tools & Integrations) and click Add new MCP server. Cursor opens ~/.cursor/mcp.json. For one project only, create .cursor/mcp.json in it instead.
Paste this and save. If the file already lists other servers, add the versyka entry inside the existing mcpServers.
```
{
  "mcpServers": {
    "versyka": {
      "url": "https://www.versyka.com/api/mcp"
    }
  }
}
```
Back in Settings → MCP, versyka shows a green dot and its 7 tools. If it doesn't, turn it off and on, or restart Cursor.
Ask in the Agent chat. Cursor shows each Versyka call and asks you to run it (or runs it automatically if you allowed that).

VS Code (GitHub Copilot)

You need VS Code with GitHub Copilot Chat.
Create .vscode/mcp.json in your project with this content. (Or run MCP: Add Server from the Command Palette, choose HTTP, paste the server address and name it versyka.)
```
{
  "servers": {
    "versyka": {
      "type": "http",
      "url": "https://www.versyka.com/api/mcp"
    }
  }
}
```
In mcp.json, click Start above the versyka entry. It shows Running with 7 tools.
Open Copilot Chat, switch to Agent mode and check the tools button lists Versyka. Then ask.

Claude Desktop and claude.ai

Open Settings → Connectors and click Add custom connector.
Name it Versyka and paste the server address, then click Add:
```
https://www.versyka.com/api/mcp
```
In a chat, open the tools menu (the + or sliders button) and make sure Versyka is turned on.
Ask. Claude shows when it uses a Versyka tool. On Team and Enterprise plans, an owner may need to add the connector for the organization first.

Other assistants that support remote MCP servers (Windsurf, Zed, Cline and others) take the same address, https://www.versyka.com/api/mcp, with the Streamable HTTP transport.

Using it

You don't call the tools yourself. Ask in plain words and the assistant picks the right tool:

Ask	What happens
"Are any of the AI models in this repo being retired?"	Scans your code and config for model IDs and lists each retired or retiring one, with its shutdown date and replacement.
"Is gpt-4-0613 still supported?"	Looks up that one model: status, shutdown date, replacement and the vendor's source.
"Check my requirements.txt for outdated API SDKs."	Compares each API SDK with the latest release on PyPI or npm and links the release notes and a migration guide.
"Upgrade this file from openai 0.28 to 1.x. Use the Versyka migration guide."	Reads the step-by-step guide, with before and after code, and applies it to your file.
"Which Anthropic models retire in the next 90 days?"	Reads the retirement calendar, filtered by vendor and date.

Want these fixes done for you as tested pull requests instead? That is what Versyka plans do.

Tools

Tool	What it does
`check_dependencies`	Check a dependency manifest (package.json, requirements.txt, pyproject.toml or Pipfile) for third-party API SDKs that are major versions behind, deprecated or replaced, against the live npm registry and PyPI.
`find_retired_models`	Scan source and config files for AI model IDs that the vendor (OpenAI, Anthropic, Google Gemini, Mistral, Cohere) has retired or scheduled for retirement.
`lookup_model`	Is this AI model ID retired or scheduled for retirement? Accepts plain IDs and gateway spellings (e.g. openai/gpt-4-0613, anthropic.claude-3-sonnet-20240229-v1:0, claude-3-opus@20240229).
`list_model_retirements`	The AI model retirement calendar: model IDs with their shutdown dates, replacements and sources, soonest first.
`get_sdk_versions`	The latest version of an npm or PyPI package, whether it is deprecated, its release-notes link, the API vendor it belongs to and a Versyka migration guide when one exists.
`list_migration_guides`	Versyka's step-by-step SDK migration guides (OpenAI, Anthropic, Google Gemini, AWS, Stripe, Firebase, Slack, Twilio, Auth0, Pinecone), optionally filtered by package name.
`get_migration_guide`	One migration guide as Markdown: what changed, each step with before and after code, how to find every affected call, how to test, and the vendor sources it is based on.

GitHub Action

The action checks every pull request, and once a week, for two things: AI model IDs that are retired or retiring (from the retirement calendar) and API SDKs that are major versions behind (the upgrade check). It writes a summary on the run and comments on the pull request with the dates, replacements and the files and lines involved.

In your repository on GitHub, click Add file → Create new file and name it .github/workflows/versyka.yml.

Paste this and click Commit changes on your default branch:

# .github/workflows/versyka.yml
name: Versyka
on:
  pull_request:
  schedule:
    - cron: "0 9 * * 1" # every Monday, so dates that come closer are caught too
permissions:
  contents: read
  pull-requests: write # to comment on pull requests; remove to only write the job summary
jobs:
  check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
      - uses: versyka/versyka-action@v1
        with:
          fail-on: retired # none | retired | soon | behind

Open a pull request. A Versyka check runs on it, writes a summary in the Actions tab and comments on the pull request when it finds something.
Choose how strict it is with fail-on (see Action options): retired fails only on models that no longer work, soon also on models retiring within 90 days, none never fails.

On GitHub Marketplace. Source and issues: github.com/versyka/versyka-action.

Action options

Input	Default	What it does
`fail-on`	retired	When to fail the check: none, retired (a model that no longer works), soon (retiring within days), or behind (also an SDK major versions behind or deprecated).
`days`	90	How many days ahead counts as "soon".
`comment`	true	Comment on the pull request. The comment is updated in place on every push, not repeated.
`path`	.	The folder to check, for monorepos.
`github-token`	github.token	Token used to comment. The default workflow token is enough.

Outputs: retired, soon and behind, the number of findings of each kind, for later steps.

What is sent where

GitHub Action: your source code never leaves the runner. The model scan runs there against the public list from /api/model-retirements. Only dependency manifests (package.json, requirements.txt, pyproject.toml, Pipfile) are sent to Versyka for the upgrade check, which reads package names and versions and stores nothing.
MCP server: only what your agent passes to a tool, for that one call. Nothing is stored or used for anything else.

More on how Versyka handles code: Security.