The deep research agent — for your agent stack, not your browser tabs.
25-step ReAct planner, 5 search & scrape tools, GTM-tuned reasoning (buying signals, ICP fit, tech stack). Drop into Claude Code, Cursor, or any MCP client — sourced, schema-valid JSON in one call instead of wiring Tavily + an LLM.
- 25-step ReAct planner
- 5 search & scrape tools
- GTM-tuned (ICP, buying signals)
- Steps per query
- 25
- Research tools
- 5
- Per query (Scale)
- €0.024
ReAct planner · configurable (8 / 25 / 40)
WebSearch · DeepSearch · NewsSearch · Scrape · Map
Pay-as-you-go from €49/mo · no minimum
Features built for agents
Every feature is designed for programmatic use — REST, MCP, structured outputs. No UI click required.
GTM-tuned ReAct planner
Buying signals, ICP detection, tech-stack inference baked into the system prompt. Not a generic chatbot — an agent that knows where sales data lives.
5 tools, picked autonomously
WebSearch, DeepSearch (search + scrape), NewsSearch (last 30 days), Scrape, Map. The agent picks which tool closes the next gap, every step.
Native MCP — for your agent stack
Streamable HTTP MCP server. Drop into Claude Code, Cursor, LangChain, any MCP client. No wrapper code.
Schema-validated outputs
JSON schema per request. The agent iterates until every required field is filled, then formats to your schema. Drop straight into pipelines.
Bulk runs + 100+ models
Hundreds of prospects in parallel. Async queue with concurrency cap, cancellation, webhook callbacks. OpenRouter under the hood — Gemini, GPT, Claude, Llama.
BYOK & multi-tenant
Override OpenRouter and Serper keys per request via headers. Clean for multi-workspace setups and unlimited volume on your own tab.
How it works
Three steps from prompt to validatable JSON.
You send a prompt
POST /research with model, prompt and optional JSON schema. Or call the research MCP tool from your agent stack.
The agent researches
WebSearch (Serper), Scrape and Map fire in parallel. A ReAct loop decides what to fetch next.
Structured JSON returns
If a schema is provided, a second LLM pass formats exactly to your schema — parseable, validatable, persist-ready.
curl -X POST https://research-agent.net/research \
-H "Content-Type: application/json" \
-H "X-Auth-Token: $RA_KEY" \
-d '{
"model": "google/gemini-2.5-flash",
"prompt": "Find Series B SaaS companies in Berlin hiring Heads of Sales",
"schema": "{\"type\":\"array\",\"items\":{\"type\":\"object\",\"properties\":{\"company\":{\"type\":\"string\"},\"role\":{\"type\":\"string\"},\"source_url\":{\"type\":\"string\"}}}}"
}' Pricing
Prices that grow with you
One credit = one autonomous research run, no matter how many sources the agent hits. 200 free credits on signup.
Starter
~€0.049 per queryFor solo builders, side projects, and first integrations.
- Autonomous ReAct research across 20+ sources
- Structured JSON outputs
- REST API + native MCP
- Email support
Growth
~€0.030 per queryFor small teams piping data into real workflows.
- Everything in Starter
- Bulk runs with concurrency cap
- Custom agent templates + schemas
- ICP scoring & buying signals
- Priority support
Scale
~€0.024 per queryBYOK for effectively unlimited volume + higher concurrency.
- Everything in Growth
- BYOK: OpenRouter & Serper via headers
- Higher concurrency limits
- Webhook callbacks
- Slack/Discord support
All plans start with 200 free credits. No credit card. Cancel anytime.
Productive in 60 seconds
Start with 200 free credits. No credit card. Full MCP and REST API.