Livy
Resolver

Livy Resolver Skill

Use this skill when a user wants an agent to use the hosted Livy Resolver MCP endpoint at https://resolver.api.livylabs.xyz/ to fetch an exact source URL. It explains how to connect, handle the OAuth authorization prompt, call the `fetch_source` tool, and tell the user what to ask so Resolver is used.

Livy Resolver MCP

Use the hosted Livy Resolver MCP service as a remote MCP server.

Primary endpoint:

https://resolver.api.livylabs.xyz/

If an MCP client requires an explicit MCP path and root discovery does not work, try:

https://resolver.api.livylabs.xyz/mcp

Do not treat this skill as infrastructure setup. The agent is expected to connect to the hosted MCP endpoint, authorize with OAuth when prompted, then use the exposed MCP tool.

Connect

Add Resolver as a remote/Streamable HTTP MCP server. Client config shapes vary, but the important value is the hosted URL:

{
  "mcpServers": {
    "livy-resolver": {
      "type": "streamable-http",
      "url": "https://resolver.api.livylabs.xyz/"
    }
  }
}

For Codex, add the hosted server and complete OAuth:

codex mcp add livy-resolver --url https://resolver.api.livylabs.xyz/
codex mcp login livy-resolver

For Claude Code, add the hosted server, then open /mcp in a Claude Code session to finish authorization:

claude mcp add --transport http livy-resolver https://resolver.api.livylabs.xyz/

For clients that do not support remote MCP servers directly, use mcp-remote:

{
  "mcpServers": {
    "livy-resolver": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://resolver.api.livylabs.xyz/mcp"]
    }
  }
}

If the client supports OAuth settings, set OAuth/auth to enabled or automatic. Do not invent OAuth issuer URLs, client secrets, access tokens, Spider keys, or Livy backend keys.

OAuth Handling

The Resolver endpoint may ask for OAuth before tools are available. OAuth is handled by the MCP client/user, not by passing credentials to the MCP tool.

When OAuth is required:

  1. Tell the user to approve/sign in through the MCP client's OAuth prompt or browser window.
  2. Use the account that has access to Livy Resolver.
  3. Let the MCP client store the access/refresh token. Do not ask the user to paste raw tokens into chat.
  4. After authorization succeeds, reconnect or refresh the MCP server.
  5. List tools or continue once fetch_source is available.

If OAuth fails, ask the user to re-authorize and check that they used the right account for https://resolver.api.livylabs.xyz/. Do not bypass OAuth, scrape the endpoint, or call raw HTTP with hand-copied bearer tokens unless the user explicitly asks for debugging.

Tool To Use

Use the fetch_source MCP tool.

Input:

{
  "url": "https://example.com/article"
}

Pass the exact URL from the user. Do not search first. Do not replace it with another URL. Do not canonicalize it unless the user explicitly asks.

The response may include receipt/provenance metadata such as:

  • receipt_id
  • status
  • fetch_elapsed_ms
  • content_bytes
  • explorer_url
  • managed_publication.status
  • registry_refs

Answer using the fetched source content. Include receipt/provenance fields when the user asks for traceability, proof, provenance, or audit details.

When To Use Resolver

Use Resolver when the user gives an exact URL and says or implies it is the source to use. Trigger phrases include:

  • source: <url>
  • only take this source
  • use this source of truth
  • fetch this exact URL
  • use Resolver
  • use Livy Resolver MCP
  • include the receipt_id
  • show provenance

If the user asks for current information without giving a URL, Resolver is not the discovery step. Ask for the source URL or use another approved tool to discover a URL first. Once an exact source URL is known, call Resolver.

User Prompts That Should Work

Use Livy Resolver MCP to fetch source: https://example.com/article
Use https://resolver.api.livylabs.xyz/ and fetch this exact URL:
https://example.com/report
Only take this source of truth and summarize it:
https://example.com/post
Connect to the Livy Resolver MCP endpoint. If OAuth appears, I will sign in. Then fetch:
https://example.com/article
Fetch this exact URL through Resolver and include the receipt_id in your answer.
Use Resolver provenance for this source URL and tell me the explorer URL, managed publication status, and registry refs if returned.

Do Not

  • Do not start the local Rust server.
  • Do not configure Spider or Livy backend secrets.
  • Do not deploy infrastructure.
  • Do not call /fetch, /crawl, /search, or other HTTP product routes unless the user specifically asks for the HTTP API instead of MCP.
  • Do not ask the user to paste OAuth access tokens into the conversation.