AI Answer to a Question (Grounded in Live Web Search)

omnial/answer

Given a natural-language question, returns a synthesized text answer with inline citation markers (e.g. [1][2]), grounded in a live web search, plus a citation list carrying each source's publish date when it has one (no links or titles).

activeper callv3

Price

Up to $0.0166 per call; you pay what the call used

You pay the price shown. If a call turns out to cost more than its quote, you pay the actual price, capped at twice the quote. No monthly minimum. Top-ups start at $5.

Data source

Exa

Data from Exa.

Use it

Five steps: search, inspect, learn, run, settle. No per-tool setup; your agent reaches this tool through the meta-tools it already has.

Connect once and your AI can find and run this tool, and any other it needs.

Install it Create an account

agent

// 1. search: find the tool
omnial_search({ query: "what you need" })

// 2. inspect: price and input schema
omnial_inspect({ tool: "omnial/answer" })

// 3. learn: read the usage guide at runtime
omnial_learn({ tool: "omnial/answer" })

// 4. run: price the exact call first, then run it
omnial_execute({ tool: "omnial/answer", input: { /* see input_schema */ }, dry_run: true })
omnial_execute({ tool: "omnial/answer", input: { /* ... */ } })

// 5. settle: the call is charged when it finishes, nothing to do

Not connected yet? Point your MCP client at https://mcp.eden-engine.com/mcp, then copy a config for your client.

Pricing

Every charge this tool can incur.

Charge	Price
Per call	$0.0166

Updated: Oct 18, 2026

Input and output schemas

Input schema

input_schema.json

{
  "type": "object",
  "required": [
    "query"
  ],
  "properties": {
    "query": {
      "type": "string",
      "description": "Question to answer (grounded in live web search)."
    }
  },
  "additionalProperties": false
}

Output schema

output_schema.json

{
  "type": "object",
  "required": [
    "answer"
  ],
  "properties": {
    "answer": {
      "type": "string",
      "description": "The synthesized answer text, with bracketed citation markers (e.g. '[1][2][3]') referencing entries in `citations` by position. Observed non-empty on both a well-known factual question and a nonsensical, ungrounded query -- the endpoint returns its best-effort synthesis (e.g. explaining a query is a 'keyboard smash') rather than an empty/error result when it cannot find a direct answer."
    },
    "citations": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "author": {
            "type": "string",
            "description": "The cited article's published byline name, when the source publishes one. Uncommon (e.g. one of fifteen citations across two queries) -- not guaranteed present."
          },
          "publishedDate": {
            "type": "string",
            "description": "ISO 8601 publish date/time of the cited page, when the source publishes one. Absent on some citations (e.g. reference/encyclopedia pages with no byline date) -- not guaranteed present on every citation."
          }
        },
        "additionalProperties": false
      },
      "description": "The sources behind the answer, typically 7-8 entries. Each citation's own source URL, page title, and any image/favicon are NOT exposed by this tool (see each field's own description below) -- only publication date and, when the source publishes one, an author byline are shipped."
    }
  },
  "description": "The answer record.",
  "additionalProperties": false
}