# Hand off a research question for a cited report

> Hand off a research question for a cited report: 3 companies fully do it, 12 with a limit, 19 carry the tag. Every listing in Web search APIs with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/search.deep-research/
- Markdown: https://www.anchorterminal.com/niches/search.deep-research/index.md (~1,750 tokens)
- Slim: https://www.anchorterminal.com/niches/search.deep-research/index.min.md (~330 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/search.deep-research/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Two or three companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 3 | 12 | 19 | 10 | 6 |

## What counts as done

A report or answer with inline citations, delivered by an async task id or a streamed response, with the cost known before the run.

Measured, for our own tests: the cited URLs resolve, and a streamed response does not time out the MCP client.

How agents ask for it: research X thoroughly and write it up; compare these companies with sources; detailed look on X

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Liner Search MCP (Liner)**: Deep Research Agent streams SSE with inline citation chunks at $200 per 1,000 runs, but there is no task id to poll, so an MCP client must hold the stream open.
- **Linkup (Linkup Technologies SAS)**: POST /research returns a task id at once, GET /research/:id polls status, reasoningDepth sets a fixed price per run, and 402 carries x402 details.
- **You.com APIs (SuSea, Inc., d/b/a you.com)**: Research API with background true returns a task_id, GET polls status and a stream endpoint sends SSE progress; prices by research_effort are listed.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Grounding with Bing Search (Microsoft Corporation)**: One Responses run returns a grounded answer with url_citation annotations, synchronously, at $14 per 1,000 tool calls plus model tokens; no async research task or report mode.
- **Brave Search API + MCP (Brave Software, Inc.)**: An Answers service with citations is listed, but it answers synchronously with no task id, and the Summarizer it replaced returned 'No web results found' on free keys until a fix.
- **Exa API + MCP (Exa Labs Inc.)**: Agent runs return an id, poll or stream SSE events, and output.grounding carries citations; price is by effort, but no budget field appears in the quickstart read.
- **Grep AI (Parcha Labs, Inc.)**: Runs are async with a run id, timeline, webhook_url, Idempotency-Key and max_cost_usd, but the cost floor depends on effort ($0.50 low, $2.50 medium, $5 high per the listing) and API access is sold on Pro and Ultra plans.
- **NewsCatcher (NewsCatcher, Inc.)**: Agent runs research events or companies in the background and return a table, not a cited prose report, at 1,000 to 6,000 credits a run.
- **Parallel Search and Task APIs (Parallel Web Systems Inc.)**: Task runs have a run_id, completed or failed status with errors, webhooks and SSE, with fixed prices by processor, but the SDKs retry creation on 429 with no idempotency key (listing weakness).
- **Perplexity Search API (Perplexity AI, Inc.)**: The Agent API's web_search tool writes a cited answer synchronously; the MCP perplexity_research tool buffered the whole stream and timed out clients until a fix in July 2026.
- **SerpKite (Anastasia Klimova, trading as SerpKite)**: /v1/research returns 202 and is polled at GET /v1/research/{id}, with fixed prices of 30, 75 or 200 credits, but the vendor's record is six weeks of status monitoring.
- **SpecStream (SpecStream SAS)**: Deep Search builds rows with a source page per value after a quoted, approved price and an idempotency key; it is a list builder, not a cited prose report.
- **Tavily API + MCP (AlphaAI Technologies Inc. dba Tavily)**: /research is async with a request_id and status, a stream option and model mini, pro or auto at 4 to 250 credits, but over MCP it hung forever when model was unspecified.
- **TinyFish Search (Tiny Fish, Inc.)**: Research runs stream SSE events (sources_searched, source_fetched, final_result with citations) with a reconnectable events endpoint and cancel, but no price or cost cap appears in the reference and deep runs take 5 to 45 minutes.
- **Valyu (Valyu.Network Ltd)**: DeepResearch creates a task with deepresearch_id to poll or a webhook, fixed credits per mode from 8 to 2,500 and hard time limits, with citations carrying passages; but task creation has no idempotency key (listing weakness).

- not this job: Web Search Tool on Amazon Bedrock AgentCore (Amazon Web Services, Inc. (regional AWS entities by account location)): Returns ranked results in one call. Cited answers are a separate Web Answer connector, not this listing.
- not this job: Arcmira (Arcmira, Inc.): No research or report endpoint; an investment research workflow is a guide for the agent, not a hosted task.
- not this job: Cloudflare Web Search API (Cloudflare, Inc.): No research or answer mode; the response is a results array.
- not this job: DeepSeek Search and Fetch API (Hangzhou DeepSeek Artificial Intelligence Co., Ltd.): Two endpoints, search and fetch; no research or answer mode is listed.
- not this job: Fetch (MCP reference server) (Model Context Protocol, a Series of LF Projects, LLC): No research mode.
- not this job: Kagi Search API (Kagi Inc.): No research or answer endpoint in the OpenAPI document.
- not this job: Mojeek Web Search API (Mojeek Limited): No research or answer mode.
- not this job: SearchAPI.io (SearchApi, LLC): No research or answer mode.
- not this job: SerpApi (SerpApi, LLC): No research mode; Google AI Mode results are scraped, not written.
- not this job: Serper (Serper): No research or answer mode.
- not this job: SERPHouse (Patoliya Infotech LLP): No research or answer mode.

## What stops agents finishing

- research tool blocks for minutes and the MCP client times out (Perplexity, Tavily)
- no idempotency so a retry starts and bills a second task
- minimum spend per run of $0.50 or more (Grep AI)

## Around it

Before: search.ground-fact. After: document-generation.report, memory.store.

## Where this comes from

Weight 10 of 100 in Web search APIs: 6 of 44 accounts: a platform running 10-plus searches and 100-plus fetches a job, three MCP issues where research tools hung or timed out, Kagi and Perplexity research tool reports

As data: https://www.anchorterminal.com/api/v1/use-cases/search.deep-research.json, or the MCP tool get_use_case with id search.deep-research. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
