MCP server

​

Prerequisites

• A Pinecone API key with access to your Nexus project. Create one in the Pinecone console.
• The MCP endpoint. Your deployment’s workspace host with the /mcp path (the base of nexus_default_workspace_data_console_url from the install output).
• Node.js, so the mcp-remote bridge can run through npx.

​

Connect

{
  "mcpServers": {
    "nexus": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://YOUR_WORKSPACE_HOST/mcp",
        "--header", "X-Pinecone-Api-Key:YOUR_PINECONE_API_KEY",
        "--transport", "http-only"
      ]
    }
  }
}

​

Tools

• list_contexts: list the contexts available to your project (each has a slug, and only curated contexts are queryable). Takes no arguments and returns an array of context objects, each shaped { slug, name, id, description, queryable, is_curating, is_optimizing, last_curated_at, last_optimized_at }. queryable turns true once the context has been curated, is_curating and is_optimizing flag whether work is running now, and the two timestamps are ISO strings or null.
• query_context: submit a question scoped to one context. Args: context (slug), question. Runs in background mode and returns a query_id immediately without waiting for the answer. The return object is { query_id, status, model }. Pass the query_id to get_query to fetch the result.
• get_query: poll the result of a submitted query. Args: query_id. Returns { query_id, answer, output_json, citations, model, status, error }. answer is the grounded answer text, or null until it is ready. output_json holds structured output when the query requested a shape, otherwise null. citations is an array of source references and error is set only when the query failed. Keep polling until status is completed, failed, or cancelled.