MCP Integration

MCP is useful when an existing assistant or agent runtime already knows how to connect to MCP servers and you want Valmar as one more tool.

Add the Valmar MCP server

This is the minimum configuration shape for an MCP-capable client. Use the base URL from your customer deployment.

{
  "mcpServers": {
    "valmar": {
      "url": "https://valmar.your-company.com/mcp",
      "headers": {
        "Authorization": "Bearer valmr_proj_sk_..."
      }
    }
  }
}

Use MCP when

• You already have an MCP-based assistant runtime.
• You want a fallback path without adding a language SDK first.
• Your assistant should keep control and call Valmar only when needed.

Available tools

• search_context searches approved project context.
• read_context retrieves one full resource from a search hit.
• request_context opens a routed request when context is missing.
• get_context_request checks the status and result of a request.
• list_recent_context_requests lists recently completed requests to avoid duplicate asks.

Unstructured search_context hits and read_context resources include optional metadata with expert_names and approved_at, so an agent can cite who contributed a saved answer and when it became available.