the `Search` tool needs some improvement

[tybalex]

I've tried to use elasticsearch MCP with Cursor as client, and I think the search tool needs better description (especially for the queryBody arg). I ingested a few text files in my es index, and asked my agent to:

find main characters of `reunion under sky` story, search in my Elasticsearch indexes

Claude4.0 was able to figure out the correct search query after 5 tries, and gpt4.1 got completely lost.
the search method should just be straightforward enough instead of the current form.

BTW here is my cursor setup:

    "Elasticsearch": {
      "command": "npx",
      "args": [
        "-y",
        "@elastic/mcp-server-elasticsearch"
      ],
      "env": {
        "ES_URL": "https://my-elasticsearch-project-xxxx.es.us-central1.gcp.elastic.cloud:443",
        "ES_API_KEY": "xxxx"
      }
    }

[swallez]

obot-platform/obot#3732 states that "a document was uploaded to Elasticsearch", a PDF file. In order to understand the problem:

• What its the format of the index?
• what is the correct search query Claude finally came up with?

Without additional information we cannot help. Thanks.

[tybalex]

---

Failed search call from gpt4.1. it got completely lost.

{
  "index": "reunionunderthestar"
}

---

GPT5 failed with

{
  "index": "reunionunderthestar",
  "explain": false,
  "profile": false,
  "query": {
    "match_all": {}
  },
  "size": 5
}

then it fixed it with

{
  "index": "reunionunderthestar",
  "explain": false,
  "profile": false,
  "queryBody": {
    "query": {
      "match_all": {}
    },
    "size": 5
  }
}

From what I can tell, the queryBody arg is kinda confusing to the LLMs

---

tried again with Claude, this time claude4 one shot it. the correct search query from Claude:

{
  "index": "reunionunderthestar",
  "queryBody": {
    "query": {
      "match_all": {}
    },
    "size": 10
  }
}

---

The reunionunderthestar index is structured with the following mapping:

Meta

• _meta.created_by: "file-data-visualizer" — indicates how the index was created.

Top-level fields

• attachment (object)

  • content – text field, with a keyword subfield (max length 256).
  • content_length – long (stores the size of the attachment).
  • content_type – text with a keyword subfield.
  • date – date field for attachment timestamps.
  • format – text with a keyword subfield.
  • language – text with a keyword subfield.

• content – semantic_text type with inference_id set to .elser-2-elasticsearch, meaning it’s configured for semantic search using Elasticsearch’s ELSER v2 model.

raw data:

{
  "isError": false,
  "content": [
    {
      "type": "text",
      "text": "Mappings for index: reunionunderthestar"
    },
    {
      "type": "text",
      "text": "Mappings for index reunionunderthestar: {\n  \"_meta\": {\n    \"created_by\": \"file-data-visualizer\"\n  },\n  \"properties\": {\n    \"attachment\": {\n      \"properties\": {\n        \"content\": {\n          \"type\": \"text\",\n          \"fields\": {\n            \"keyword\": {\n              \"type\": \"keyword\",\n              \"ignore_above\": 256\n            }\n          }\n        },\n        \"content_length\": {\n          \"type\": \"long\"\n        },\n        \"content_type\": {\n          \"type\": \"text\",\n          \"fields\": {\n            \"keyword\": {\n              \"type\": \"keyword\",\n              \"ignore_above\": 256\n            }\n          }\n        },\n        \"date\": {\n          \"type\": \"date\"\n        },\n        \"format\": {\n          \"type\": \"text\",\n          \"fields\": {\n            \"keyword\": {\n              \"type\": \"keyword\",\n              \"ignore_above\": 256\n            }\n          }\n        },\n        \"language\": {\n          \"type\": \"text\",\n          \"fields\": {\n            \"keyword\": {\n              \"type\": \"keyword\",\n              \"ignore_above\": 256\n            }\n          }\n        }\n      }\n    },\n    \"content\": {\n      \"type\": \"semantic_text\",\n      \"inference_id\": \".elser-2-elasticsearch\"\n    }\n  }\n}"
    }
  ]
}