Skip to content

Exa · Multi-step research, list building, enrichment

Research run

Exa agent run (multi-step research, list building, enrichment), by Exa. Create an asynchronous Agent run. Asynchronous: returns a run id; poll exa/agent-run-get for the output, then GET /v1/runs/{run_id} shows the final price. Runs end within one hour. Price: Fixed effort: minimal $0.012, low $0.025, medium $0.10, high $0.50, xhigh $1.00 per run; auto and max metered ($0.10 per compute unit, $0.005 per search, $0.02 per email, $0.07 per phone number) up to budget.maxCostDollars (default $5, $20 for max). Billed at the cost Exa reports for the call (costDollars.total); the quote is the upper bound.

Usage based ExaResearchexa/agent-runListed

Activation pending; your agent will see it as unavailable until then.

Use this tool with your agent

Give this to your agent
Use my connected Agentik MCP server. I want to use the tool exa/agent-run (Exa agent run (multi-step research, list building, enrichment)). Call inspect_tool with {"tool_id": "exa/agent-run"} first and tell me the price before anything runs. Ask me for the inputs it needs; never invent values, IDs or file URLs. When I confirm, call run_tool with tool_id "exa/agent-run" and my inputs, follow it with get_run if it is still running, and give me the result and the price Agentik reports. If the tool is unavailable, say so and use discover_tools to propose an alternative.
Show the prompt

Paste it into Claude, ChatGPT or Codex: your agent takes it from there, and shows the price before it runs.

Parameters

NameTypeRequiredDescription
AcceptstringNoSet to `text/event-stream` to receive server-sent events. Options: application/json, text/event-stream.
Exa-BetastringNoComma-separated beta feature tokens for opting into experimental features.
querystringYesNatural-language question or instructions for the request.
systemPromptstringNoAdditional instructions that guide generated output or agent behavior. Use this for source preferences, novelty constraints, duplication constraints, or other behavior guidance.
inputobjectNoRecords to process and records or entities to exclude from the answer.
outputSchemaobject | nullNo
effortstringNoCost and reasoning effort preference for the run. `auto` lets Exa choose the appropriate effort. `max` is the highest-effort public beta tier for work where completeness and thoroughness matter more than latency or cost, including large list building, deep multi-source research, and criteria that are hard to verify. Options: minimal, low, medium, high, xhigh, auto, max.
previousRunIdstringNoAgent run ID. New run IDs are returned with the `agent_run_` prefix.
metadataobjectNoCaller-provided metadata stored with the run.
dataSourcesobject[]NoExa Connect data providers to enable for the run. Each entry enables all of that provider's tools.
budgetobjectNoOptional per-run spending limit for the metered `auto` and `max` efforts. Runs that finish early may cost less than the limit.

How your agent calls it

First inspect_tool for the current schema and price, then run_tool with the published example after confirming your inputs:

{
  "tool_id": "exa/agent-run",
  "input": {
    "query": "In one sentence, what is the Model Context Protocol?",
    "effort": "minimal"
  }
}

Price and billing

Price
Usage based
Model
Usage based
Detail
Billed at the cost the provider reports for the call, never above $100. Your agent sees the maximum before running.
  • The provider’s price, 0% markup.
  • One prepaid balance for every provider; your agent sees the price before it runs.
  • Pay for measured usage; unused reservations are released.