deepseek_research
Research any topic across multiple web sources and return a compact cited report in one call, with numbered citations matched to sources instead of raw pages filling your context.
Instructions
Research a topic across the web and get a cited report in one call.
The server searches several angles, reads 6-24 sources and returns a compact report whose [n] citations match the ids in the sources array, instead of filling your context with raw pages. About $0.003 (depth "low") to $0.03 ("max"). depth: "low" quick overview, "medium" (default) most questions, "high" decisions with several viewpoints, "max" broad or contested topics. sources="academic" favours peer-reviewed papers, "technical" official docs, specs and repos. Numbers missing from their cited source are marked [unverified] (unverified="drop" removes those sentences). If status is "running", call deepseek_check_research with the job_id; do not start the same research again. With synthesis="client" you get the ranked sources in pages and write the report yourself; the server then verifies its citations and numbers. At most 3 jobs run at once.
When NOT to use: one fact or query (deepseek_search), several independent queries (deepseek_search_many), known URLs (deepseek_fetch), or fields as JSON (deepseek_extract).
Returns: {"ok": True, "status": "done"|"running"|"awaiting_report", "job_id"?, "report": str, "sources": [{id, title, url, date, doi, pmid, pmcid, access, source_type, authors}], "unverified_numbers": [...], "cost_usd": float}.
Tool call template
Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.
{
"id": 1,
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"arguments": {
"topic": "<topic>"
},
"name": "deepseek_research"
}
}Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| lang | No | Language code such as 'en' or 'pt-BR' to restrict search sources to (default None) | |
| depth | No | Depth tier controlling queries, sources and reasoning (default 'medium') | medium |
| topic | Yes | Complex research topic or question to investigate (non-empty, max 500 chars) | |
| refresh | No | Bypass local cache and perform fresh research pipeline (default False) | |
| sources | No | Source preference: 'any' (default), 'academic' (papers) or 'technical' (docs, specs, repos) | any |
| synthesis | No | Who writes the report: 'server' (default) or 'client' (you get ranked sources, write it, server verifies it) | server |
| unverified | No | Sentences with numbers not found in their sources: 'flag' marks them (default), 'drop' removes them | flag |
| strip_boilerplate | No | Remove boilerplate from fetched source pages before synthesis (default True) |
Output Schema
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||