Agents

Use Viv with your agent

Viv has an MCP server. Add one URL to your agent and it can search the site's programs and modules, read a post, and read the language reference. Every tool only reads, so there is no account and no sign-in. To run a program, paste it into the workbench or build Viv from the repository.

The endpoint

https://viv-lang.org/mcp

It speaks the Streamable HTTP transport, revisions 2025-03-26 through 2026-07-28. Most clients need only this URL. Pick yours below and copy the line.

Claude Code

claude mcp add --transport http viv https://viv-lang.org/mcp

Then type /mcp in a session to see the server and its tools. Claude Code's MCP docs cover the rest.

Claude (claude.ai and Desktop)

Open Customize, then Connectors, then Add custom connector.
Paste the URL and leave the OAuth fields under Advanced settings empty.

https://viv-lang.org/mcp

On Team and Enterprise plans an owner adds the connector in Organization settings. A free account gets one custom connector. Anthropic's help article has the details.

Cursor

Add the server to Cursor's mcp.json:

{
  "mcpServers": {
    "viv": {
      "url": "https://viv-lang.org/mcp"
    }
  }
}

The file is .cursor/mcp.json in a project, or ~/.cursor/mcp.json for every project. Cursor's MCP docs have the rest.

VS Code (Copilot agent mode)

Add the server to VS Code's mcp.json:

{
  "servers": {
    "viv": {
      "type": "http",
      "url": "https://viv-lang.org/mcp"
    }
  }
}

The file is .vscode/mcp.json in a workspace; the command MCP: Open User Configuration opens the one for every workspace. VS Code's MCP docs have the rest.

ChatGPT

On the web, open the Plugins page, press the plus button and choose Add custom MCP server.
Give it a name, enter this URL under Connection, choose no authentication, and create it as a plugin.

https://viv-lang.org/mcp

OpenAI's page on connecting an MCP server to ChatGPT has these steps and says which account and workspace policies apply.

Without MCP

An agent that cannot use MCP can still read the language. The prompt pack is the grammar, the Python habits that fail and every documented example with its output, about 33k tokens. Paste it into the conversation, or give the agent the link:

https://viv-lang.org/viv-prompt-pack.md

The full reference adds the command line, the backend matrix and the engines that run each builtin:

https://viv-lang.org/llms-full.txt

Both are plain files: viv-prompt-pack.md and llms-full.txt. The MCP server serves the same two as the resources viv://prompt-pack and viv://llms-full.

What to ask

One prompt per tool, to try once the server is connected. The agent picks the tool.

`search`

Searches the published posts, the example programs, the lib/ modules, the builtins and the documented examples.

Search Viv for a module that solves a symmetric eigenvalue problem, and show me how to import it.

`get_post`

Returns a published post: its source, its tests, its lane and the other versions in its family.

Get the Viv post runge-phenomenon and explain what its tests check.

`viv_reference`

Returns the language reference: the grammar, the Python habits that fail, the builtins, the examples by topic.

Read the Viv reference, then write a Viv program that integrates sin from 0 to pi with the trapezoid rule and tests the answer.

Limits

Requests are rate limited per address.
The posts the tools see are the ones published when the site was last built.
A solution post is never returned or searched, so a problem's answers stay hidden.

The client steps above come from each vendor's documentation as read on 2026-09-24, with Cursor's, VS Code's and ChatGPT's read again on 2026-10-16. The server was tested with direct requests for every tool; none of these clients has been tried against it yet.