Munition
Docs menu

Web research API

Reference for Munition's web research routes, Exa, Firecrawl and Serper search and page reading: routes, input, output, price and limits.

Ten operations from three providers: Exa (neural search, page contents, similar pages, cited answers), Firecrawl (pages rendered with JavaScript, and web search) and Serper (Google web, news, images and shopping). Each one is sold three ways at one price: an x402 route, its MPP mirror under /v1/mpp/, and an MCP tool paid from the account balance. The MPP route and the MCP tool take the same input and return the same result as the x402 route.

How these routes work

  • One request, one price. Every route is a POST with a JSON body and a JSON result, at a fixed price per call whatever the number of results. The tables below give each route, its tool id in GET /v1/tools/prices and its price in USDC.
  • Invalid input is never charged. An unpaid request gets its challenge (402) before the body is read, on the x402 and MPP routes. A request with a payment and an invalid body gets 400 invalid_input before the payment is verified.
  • Timeouts. Munition waits up to 25 seconds for Exa and Serper and 28 seconds for Firecrawl.
  • Extra fields. Result items can carry more fields from the provider than the tables list.
  • Failures after payment keep the payment and return 422 or 502 upstream_failed with paidCallId. Send the same request again with the same Idempotency-Key to call the provider again without a second charge: see Errors and refunds.

Exa

Neural web search: up to 25 ranked results, with page contents in the same call if you ask for them.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/exa/search exa_search_direct 0.01
MPP POST /v1/mpp/exa/search exa_search_mpp 0.01
Account balance MCP tool web_search exa_search 0.01

Body of POST /v1/exa/search:

Field Type Required Default Limits and notes
query string yes — 1 to 2,000 characters, in natural language, not counting spaces at either end.
numResults integer no 10 1 to 25.
type string no auto auto, neural, keyword or fast.
category string no — company, research paper, news, pdf, github, tweet, personal site, linkedin profile, financial report or people.
includeDomains array of strings no — Up to 50 domains to search in.
excludeDomains array of strings no — Up to 50 domains to leave out.
startPublishedDate string no — An ISO 8601 date or date-time (2026-01-31 or 2026-01-31T00:00:00Z), passed to Exa as the earliest publication date. Anything else is refused with 400 before payment.
endPublishedDate string no — An ISO 8601 date or date-time, passed to Exa as the latest publication date. Anything else is refused with 400 before payment.
contents object no — The page contents to include in each result, below. Omit it for metadata only.

The contents object of POST /v1/exa/search and POST /v1/exa/find-similar:

Field Type Required Default Limits and notes
text boolean or object no — true for the page text, or an object with maxCharacters (up to 50,000).
highlights boolean or object no — true, or an object with numSentences (up to 20) and highlightsPerUrl (up to 10).
summary boolean or object no — true, or an object with a query (up to 500 characters) to focus the summary.

Response of POST /v1/exa/search:

Field Type Always present Notes
query string yes The query you sent.
results array of objects yes Each result has url, and when known title, publishedDate, author and score, plus text, highlights and summary when contents asked for them.
searchType string no The search type Exa used.
meta object yes resultCount and searchedAt.

Page contents

The clean text of up to 20 known URLs from Exa's index, with a live crawl when needed. The cheapest way to read many pages.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/exa/contents exa_contents_direct 0.002
MPP POST /v1/mpp/exa/contents exa_contents_mpp 0.002
Account balance MCP tool web_contents exa_contents 0.002

Body of POST /v1/exa/contents:

Field Type Required Default Limits and notes
urls array of strings yes — 1 to 20 URLs.
text boolean or object no true true, false, or an object with maxCharacters (up to 50,000).
highlights boolean or object no — true, or an object with numSentences (up to 20).
summary boolean or object no — true, or an object with a query (up to 500 characters).
livecrawl string no — never, fallback or always: when Exa crawls the page live instead of using its index.

Response of POST /v1/exa/contents:

Field Type Always present Notes
results array of objects yes One per page read: url, title, text, and highlights or summary when asked for.
meta object yes resultCount and fetchedAt.

Similar pages

Up to 25 pages similar to a URL: competitors, related articles, alternatives.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/exa/find-similar exa_find_similar_direct 0.01
MPP POST /v1/mpp/exa/find-similar exa_find_similar_mpp 0.01
Account balance MCP tool web_find_similar exa_find_similar 0.01

Body of POST /v1/exa/find-similar:

Field Type Required Default Limits and notes
url string yes — The page to find similar pages for.
numResults integer no 10 1 to 25.
excludeSourceDomain boolean no true Leave out pages from the domain of url.
includeDomains array of strings no — Up to 50 domains to search in.
excludeDomains array of strings no — Up to 50 domains to leave out.
contents object no — The same contents object as Search.

Response of POST /v1/exa/find-similar:

Field Type Always present Notes
query string yes The url you sent.
results array of objects yes The same result objects as Search.
searchType string no The search type Exa used.
meta object yes resultCount and searchedAt.

Cited answer

A direct answer to a question, grounded in a live web search, with the source pages as citations.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/exa/answer exa_answer_direct 0.01
MPP POST /v1/mpp/exa/answer exa_answer_mpp 0.01
Account balance MCP tool web_answer exa_answer 0.01

Body of POST /v1/exa/answer:

Field Type Required Default Limits and notes
query string yes — The question, 1 to 2,000 characters, not counting spaces at either end.
text boolean no false Include the full text of each citation.

Response of POST /v1/exa/answer:

Field Type Always present Notes
query string yes The question you sent.
answer string yes The answer.
citations array of objects yes The source pages, as result objects (url, title, and text when asked for).
meta object yes citationCount and answeredAt.

Firecrawl

Scrape a page

One URL rendered with JavaScript and converted to Markdown, and optionally its HTML, links or a summary. Use it when a plain fetch returns an empty page, a page that needs JavaScript, or a 403.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/firecrawl/scrape firecrawl_scrape_direct 0.015
MPP POST /v1/mpp/firecrawl/scrape firecrawl_scrape_mpp 0.015
Account balance MCP tool web_scrape firecrawl_scrape 0.015

Body of POST /v1/firecrawl/scrape:

Field Type Required Default Limits and notes
url string yes — The page to scrape.
formats array of strings no ["markdown"] 1 to 4 of markdown, html, links and summary.
onlyMainContent boolean no true Strip navigation, footers and other boilerplate.
waitFor integer no — Milliseconds to wait for dynamic content, 0 to 10,000.
timeout integer no — Milliseconds, 1,000 to 30,000.

Response of POST /v1/firecrawl/scrape:

Field Type Always present Notes
url string yes The final URL, after redirects.
title string or null yes The page title.
markdown string or null yes The page content as Markdown.
html string or null no When formats asked for it.
links array of strings no When formats asked for them.
summary string or null no When formats asked for it.
statusCode integer or null yes The HTTP status of the page.
meta object yes fetchedAt, and characters, the length of markdown.

Search with page content

A web search with up to 20 results, and optionally the full Markdown of each result in the same call.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/firecrawl/search firecrawl_search_direct 0.03
MPP POST /v1/mpp/firecrawl/search firecrawl_search_mpp 0.03
Account balance MCP tool web_search_scrape firecrawl_search 0.03

Body of POST /v1/firecrawl/search:

Field Type Required Default Limits and notes
query string yes — 1 to 1,000 characters.
limit integer no 5 Results wanted, 1 to 20.
lang string no — Language code, 2 to 5 characters, such as en or fr.
country string no — Two-letter country code, such as us or fr.
tbs string no — Time filter, up to 40 characters: qdr:d (day), qdr:w (week), qdr:m (month).
scrapeContent boolean no false Also scrape each result to Markdown.

Response of POST /v1/firecrawl/search:

Field Type Always present Notes
query string yes The query you sent.
results array of objects yes Each has url, title and description, and markdown when scrapeContent is true.
meta object yes resultCount and searchedAt.

Serper (Google)

Google results through Serper: web, news, images and shopping. The four routes take the same body:

Body of POST /v1/serper/search, POST /v1/serper/news, POST /v1/serper/images and POST /v1/serper/shopping:

Field Type Required Default Limits and notes
q string yes — The Google query, 1 to 1,000 characters, not counting spaces at either end.
gl string no us Country code, exactly 2 letters, such as us, fr or de; sent to Serper in lower case.
hl string no en Interface language, 2 to 5 characters, such as en or fr.
num integer no 10 Results wanted, 1 to 100. The answer never has more than num results.
page integer no 1 Result page, 1 to 10.
tbs string no — Time filter, up to 40 characters: qdr:d, qdr:w, qdr:m or qdr:y.

Web results

Organic results with snippets, the answer box, the knowledge graph, people also ask and related searches.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/serper/search serper_search_direct 0.004
MPP POST /v1/mpp/serper/search serper_search_mpp 0.004
Account balance MCP tool google_search serper_search 0.004

Response of POST /v1/serper/search:

Field Type Always present Notes
query string yes The query you sent.
organic array of objects yes Each has title and link, and when Google gives them snippet, position and date.
answerBox object or null no Google's answer box, null when there is none.
knowledgeGraph object or null no Google's knowledge graph panel, null when there is none.
peopleAlsoAsk array of objects no Related questions.
relatedSearches array of objects no Each has a query.
meta object yes resultCount and searchedAt.

News

Google News articles, with source and date.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/serper/news serper_news_direct 0.004
MPP POST /v1/mpp/serper/news serper_news_mpp 0.004
Account balance MCP tool google_news serper_news 0.004

Response of POST /v1/serper/news:

Field Type Always present Notes
query string yes The query you sent.
news array of objects yes Each has title and link, and when known snippet, date, source, imageUrl and position.
meta object yes resultCount and searchedAt.

Images

Google Images results, with image URLs, dimensions and source pages.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/serper/images serper_images_direct 0.004
MPP POST /v1/mpp/serper/images serper_images_mpp 0.004
Account balance MCP tool google_images serper_images 0.004

Response of POST /v1/serper/images:

Field Type Always present Notes
query string yes The query you sent.
images array of objects yes Each has imageUrl, and when known title, imageWidth, imageHeight, thumbnailUrl, source, link and position.
meta object yes resultCount and searchedAt.

Shopping

Google Shopping products, with price, seller and rating.

Paid with Route or tool Tool id Price (USDC)
x402 POST /v1/serper/shopping serper_shopping_direct 0.004
MPP POST /v1/mpp/serper/shopping serper_shopping_mpp 0.004
Account balance MCP tool google_shopping serper_shopping 0.004

Response of POST /v1/serper/shopping:

Field Type Always present Notes
query string yes The query you sent.
shopping array of objects yes Each has title, and when known source, link, price, rating, ratingCount, imageUrl and position.
meta object yes resultCount and searchedAt.

Example

An Exa search for five results. Called without payment, it answers 402 with the price in the PAYMENT-REQUIRED header; Copy-paste examples shows how to sign the payment and send the request again, in curl, TypeScript and Python.

bash
curl -sS -i https://api.munition.io/v1/exa/search \
  -H 'Content-Type: application/json' \
  --data '{"query":"x402 payment protocol for AI agents","numResults":5,"type":"auto"}'

Paid, it costs 0.01 USDC and answers 200. A response of POST /v1/exa/search looks like this:

json
{
  "query": "x402 payment protocol for AI agents",
  "results": [
    {
      "title": "x402 — HTTP-native payments",
      "url": "https://www.x402.org/",
      "publishedDate": "2025-05-06T00:00:00.000Z",
      "author": null,
      "score": 0.21
    }
  ],
  "searchType": "neural",
  "meta": { "resultCount": 1, "searchedAt": "2026-09-09T00:00:00.000Z" }
}