Perplexity Web Research MCP server

How an agent would call it

Two steps: search returns refs, then data_fetch opens one of them. The arguments below are the arguments that were sent.

{
  "tool": "search",
  "arguments": {
    "query": "How do teams measure whether AI answer engines cite their brand?",
    "sources": [
      "web-research"
    ],
    "limit": 3,
    "runId": "<from run_begin>"
  }
}

{
  "tool": "data_fetch",
  "arguments": {
    "sourceId": "web-research",
    "ref": {
      "id": "answer:How do teams measure whether AI answer engines cite their brand?"
    },
    "runId": "<from run_begin>"
  }
}

This runs against an account — send Authorization: Bearer <token> with the call. Get one at app.bedrocktap.com/dashboard/tokens; the free tier needs no card, and it reaches web-research like every other source.

What it costs. 30 credits per question asked — the most expensive call in the catalogue. Refs and fetches add nothing, so the 21 refs above cost 30 credits in total.

What came back

{
  "ref": {
    "id": "answer:How do teams measure whether AI answer engines cite their brand?",
    "title": "Researched answer"
  },
  "text": "Teams usually measure this with a **prompt set** and a **citation log**: they run the same representative questions across answer engines, record whether the brand is **mentioned** and whether its domain is **cited**, then track those results over time by query, engine, and competitor set.[13][19][20]…",
  "fetchedAt": "2026-09-12T15:04:45.667Z"
}

One search returned 21 refs: the answer, then one per citation. The answer ref carried kind, answer, model and citationCount — 20 citations behind a 1,642-character answer whose claims were numbered inline against them. Each citation ref carried its source URL as both its id and its title.

Observed 2026-09-12. One call against a live source — ask again tomorrow and the rows move.

What a row contains

The first row is the answer: no URL, and meta carrying the answer text, the model that wrote it and the citation count. Every row after it is one cited source, carrying its URL. The claims in the answer are numbered against those citations, so a reader can check which source stands behind which sentence. A citation proves a page was consulted, not that the answer is right.

A question it answers. How do teams in this market actually solve this, and what are they citing when they say so?

Routing guidance

What an agent reads to decide whether web-research is the right source, and what to reach for when it is not.

Answers

A researched answer with attribution when you do not know which sites hold it. The citations are the valuable part — treat the prose as a lead, not a fact.

Query shape

A natural-language QUESTION, not keywords.

Price

Priced in credits, per row, per query or per call depending on the source. Every call is quoted before it runs — `cost_estimate` returns the exact figure and spends nothing. See /pricing.

Legal basis

Public API — the platform’s own documented, openly available interface.

Latency

Moderate (~5-15s).

Prefer instead

`grounded-answers` when the same question can be answered for fewer credits. Come here for a second, independent engine. `google-serp` when you need actual RANKINGS rather than an answer.

Cannot answer

Search volume, keyword metrics, or anything requiring a stable citable snapshot.

Caveat

Non-deterministic: the same query returns a fresh answer every time, never a replay.

Related sources