What you get
Every article is broken into statements: one-sentence facts such as "Nvidia expects to grow at a strong pace again in 2027", each with its sentiment (bullish, bearish or neutral), importance (low, medium, high, ultra), the symbols and sector it concerns, the source URL and domain, the article's post time and, when the article has one, its image.
Statements that report the same development are clustered into topics: "30-year Treasury yield breaks 5%" holds the statements of many outlets. Topics are grouped into standing themes such as Central banks & rates or Energy. Trending lists the developments the most outlets picked up: a live list and the top of every 2-hour slot of the past week.
- History: how far back you can read depends on your plan. Every response says it (
snapshot.history_days,snapshot.window_start), andGET /v2/metagives the current counts. - Times: every statement carries
published_at(when its article was published) andindexed_at(when it entered the API). - Access: through RapidAPI, with your RapidAPI key.
This site, fintopic.news, is a front page built entirely on the same API.
Quick start
1. Get a key. Pick a plan on RapidAPI. Then
copy your X-RapidAPI-Key from the endpoint playground.
2. Make a request. The newest high-importance statements about Nvidia:
curl -G https://finance-pulse.p.rapidapi.com/v2/statements \
--data-urlencode "symbol=NVDA" \
--data-urlencode "importance_min=high" \
--data-urlencode "limit=3" \
-H "X-RapidAPI-Key: YOUR_RAPIDAPI_KEY" \
-H "X-RapidAPI-Host: finance-pulse.p.rapidapi.com"
import requests
API = "https://finance-pulse.p.rapidapi.com"
HEADERS = {"X-RapidAPI-Key": "YOUR_RAPIDAPI_KEY", "X-RapidAPI-Host": "finance-pulse.p.rapidapi.com"}
r = requests.get(f"{API}/v2/statements", headers=HEADERS,
params={"symbol": "NVDA", "importance_min": "high", "limit": 3})
for s in r.json()["data"]:
print(s["published_at"][:16], s["sentiment"], s["statement"], f"({s['source_domain']})")
const API = "https://finance-pulse.p.rapidapi.com";
const HEADERS = { "X-RapidAPI-Key": "YOUR_RAPIDAPI_KEY", "X-RapidAPI-Host": "finance-pulse.p.rapidapi.com" };
const r = await fetch(`${API}/v2/statements?symbol=NVDA&importance_min=high&limit=3`, { headers: HEADERS });
for (const s of (await r.json()).data) console.log(s.published_at.slice(0, 16), s.sentiment, s.statement);
3. Read the response. One statement, plus the snapshot block every v2 response carries:
{
"data": [
{
"id": "917490c7-c112-542e-86e3-23dacc562f76",
"statement": "Hewlett Packard Enterprise is launching a new AI factory architecture using NVIDIA DSX platforms to enhance production of advanced servers.",
"published_at": "2026-10-13T09:45:19Z",
"indexed_at": "2026-10-13T10:00:32.723720Z",
"sentiment": "bullish",
"importance": "high",
"source_type": "news",
"source_url": "https://ec.ltn.com.tw/article/breakingnews/5605064",
"source_domain": "ec.ltn.com.tw",
"article_title": "和碩導入輝達DSX AI工廠架構 強化先進伺服器量產",
"image_url": null,
"symbols": ["NVDA"],
"entities": [
{
"code": "NVDA",
"kind": "equity",
"name": "NVIDIA Corporation",
"exchange": "NASDAQ",
"country": "USA",
"match": "tagged"
}
],
"sectors": ["InformationTechnology"],
"topic_id": "6d69e7ba-337e-432b-b539-8ca1345d50f6",
"topic_name": "NVIDIA launches DSX platform for AI factory design",
"theme_id": "a649bcf3-ff7a-592c-b4d3-b32a369fd55c"
},
… 2 more
],
"next_cursor": "WzE3OTE4Nzg1NDAsIjE4OTI2OTUzLTUyNTMtNTcxNS05NDU0LThhNjMwMTM4ZDgxNSJd",
"snapshot": {
"built_at": "2026-10-13T10:01:53.282220+00:00",
"anchor": "2026-10-13T10:01:53.282220+00:00",
"window_start": "2026-08-12T10:01:53.282220+00:00",
"lag_days": 0.0
}
}
published_at is when the article was published; indexed_at is when the statement entered the API.
article_title is the article's own headline, in the article's language (not always English) and not updated if
the publisher changes it later, next to source_url, the link to it. topic_id
links the fact to its development, and next_cursor continues the list. GET /v2/meta lists every accepted
filter value.
Prefer Python, or an AI assistant? pip install finance-pulse gives you a Python client and an MCP server for
Claude and Cursor: see Use Finance Pulse from Claude, Cursor or Python.
The data model
| Level | What it is | Example | Endpoints |
|---|---|---|---|
| Statement | One extracted fact from one article | "Brent fell 3% to $81" (reuters.com, bearish, high) | /v2/statements, /v2/feed |
| Topic | One development: the statements of many outlets about the same event | "Oil prices slide to 2-week lows" | /v2/topics, /v2/trending |
| Storyline | One symbol's topics that continue each other: an event, then its follow-ups and reactions | Under MSFT: "Microsoft suspends green-card sponsorship" and what followed |
/v2/storylines |
| Theme | A standing subject that groups topics | Energy | /v2/themes |
Counts on topics and themes (statements, sources) cover everything the API holds for them, also beyond your
plan's history; total_statements is everything the topic ever held. sources always means distinct outlets (domains), not URLs.
A storyline belongs to one symbol. News about two companies sits in a storyline under each; related on a
storyline lists the other symbols' storylines that share the most topics with it. Countries have no storylines.
Every topic says which storylines it is in (storylines); see
Detecting developing stories.
Sentiment is a label on a statement. A symbol's, topic's or theme's sentiment is the count of its statements'
labels, and score is (bullish − bearish) / statements. A bearish sentence that names NVDA and AMD counts for both.
Conventions
- Paging. Lists return
next_cursor; pass it back as?cursor=until it isnull. Cursors stay valid while new data arrives, so pages never skip or repeat rows./v2/feedalways returns a cursor (see the feed guide). - Caching. Every response has an
ETag, which changes whenever the API's data is updated. Send it back asIf-None-Matchand you get an empty304until then. - Errors are
application/problem+json:{"title", "status", "detail", "param"}. Any malformed request is a400, including an unknown query parameter (a typo never silently widens your query); an id outside the window is a404. - Repeatable filters.
symbol,sector,sentimentandsource_typeaccept several values:?symbol=NVDA&symbol=AMDor?symbol=NVDA,AMD(OR).
What this data is, and what it is not
- Symbols are entities, and only equities and ETFs are tickers. Every symbol is a canonical code with a
kind: equities and ETFs under their ticker (US plain,NVDA; elsewhere with the exchange suffix,0700.HK,SAP.DE), and indices, rates, currencies, crypto, commodities, organisations and private companies under short codes (SPX,US10Y,USD,BTC,GOLD_COMMODITY,FED,OPENAI). On a statement'sentities,"match"is"tagged"for a ticker assigned to the statement and"text"for an entity named in its own words (the Fed, an index, a private company). Entities are recognised automatically and not every mention is caught, so symbol counts are conservative. Old codes still filter (MICfinds Micron'sMUstatements). See the screener guide. - Sectors are assigned per statement, from its text: the 11 GICS sectors (Energy, Financials, InformationTechnology, …). A sentence about Tesla's view of GDP can be tagged Financials.
- Coverage can have gaps. Before you compute a baseline, look at a whole sector's news by day
(
/v2/series?sector=…&source_type=news&interval=day): a run of near-empty days there is a gap in the data, not a quiet market. - Storylines are a reading aid, not a count of events. The grouping is automatic and can be wrong or
incomplete: one real story can be split over several storylines, and a storyline can hold a topic
that does not belong. Do not count storylines to count stories. A membership whose
placed_atisnullhas no recorded placement time: do not use it in a point-in-time test. indexed_atcan be long afterpublished_at. Compare the two before treatingindexed_atas the moment the news was available.- There is no backtest-ready history. Event studies on
published_atare possible but contaminated by look-ahead: a late article is filed under the hour it was published. For honest point-in-time tests, record/v2/feedfrom day one; that log is your dataset.
Versions and changes
- 2.4.1 (Oct 2026). A value that is not an id at all (a typo, a placeholder) answers 400 naming the parameter;
404 still means an id that is unknown or older than your plan's history. A ticker is no longer an alias of a
different listed company:
symbol=SEBis Seaboard's; the Swedish bank SEB's news is underSVKEF./v2/series,/v2/storylinesand/v2/symbols/{code}carry an example symbol, so they answer in RapidAPI's playground as they stand. -
2.4.0 (Oct 2026). History per plan: how far back you can read depends on your plan (see the pricing page). Every response says how many days your plan includes (
snapshot.history_days, new) and the oldest post time you can see (snapshot.window_start). Statements older than that are not returned, nor topics and storylines with nothing newer: their ids answer 404.period=windowon/v2/symbols,/v2/sectorsand/v2/symbols/{code}covers your plan's history, in whole hours. A storyline'stopics,first_seenandnameare over the topics you can read; the other counts on topics, storylines and themes cover everything the API holds. Responses areCache-Control: privateand theirETagdiffers per plan./v2/feedcan return a page with fewer thanlimitstatements while more follow: keep calling withnext_cursor. -
2.3.0 (11 Oct 2026). Additive. New level: storylines, the topics about one symbol that continue one story. New:
/v2/storylines?symbol=and/v2/storylines/{id};storylineson every topic (nullwhile a new topic is not placed yet,[]when it is in no storyline, else one entry per symbol withplaced_at);top_storylineson/v2/symbols/{code};storyline_id=on/v2/topicsand/v2/statements; storyline keys in/v2/meta./v2/metano longer returnsincidentsandbuild_interval_seconds. Storylines are per symbol: a story about two companies appears under each, linked byrelated. A topic'srevisiondoes not cover itsstorylines. -
2.2.0 (9 Oct 2026). Additive, with one change to counts. New:
/v2/symbols/{code}, one entity's page;q=to look a symbol up by name;since/untilon/v2/symbolsand/v2/sectors;mentions_single,bullish_single,bearish_single,neutral_singleandscore_singleon symbol rows; coverage figures in/v2/meta;matchon a statement's entities. Counts change: central banks, indices, private companies, rates, crypto, currencies and commodities are counted more completely from this release. Don't compare their counts with values you stored before it. - 8 Oct 2026, codes. Eight codes that meant two things now mean one. The non-company side got its own code:
GOLD_COMMODITY,WTI_COMMODITY,SOL_CRYPTO,LINK_CRYPTO,BOC_ORG,RBA_ORG,IDR_FX,INR_FX. The plain codes (GOLD,WTI,SOL,LINK,BOC,RBA,IDR,INR) are now only the listed companies. - 2.1.0 (7 Oct 2026). Additive. Every statement carries
article_title, the headline of the article it came from. A topic'slatest_statementcarriesarticle_titleandsource_urltoo, so one call to/v2/trendinggives a story, the newest publisher headline for it, the outlet, the time, the image and the link. - 2.0.0 (2 Oct 2026). The unprefixed v1 endpoints (
/statements,/topics,/themes,/trending,/symbols,/meta) are removed; everything they did is in/v2. The deprecatedindustry_groupparameter and field and/v2/industry-groupsare removed too: usesector,sectorsand/v2/sectors./healthis unchanged. - 1.4.0 (2 Oct 2026).
/v2/themestakessymbol=,sector=andsort=(size, topics, recent, velocity). Topics carrystatements_7dandrevision, which changes whenever the topic or its last-7-day statements change: compare it to skip refetching unchanged topics./v2/symbolsreturns up to 20,000 rows. - 1.3.0 (1 Oct 2026). Canonical symbols, in v1 and v2: one code per entity (Micron is always
MU, neverMIC; SpaceX isSPACEX, notSPCE), and codes that resolve to nothing are no longer listed. v2 statements, topics and themes carryentities(code, kind, name, exchange, country);/v2/symbolsrows carrykindandnameand takekind=(e.g.kind=equity,etf). Old codes keep working as filters. - 1.2.2 (1 Oct 2026).
sectorreplacesindustry_group(the values always were the 11 GICS sectors): responses carrysectors, filters takesector=, and/v2/sectorsreplaces/v2/industry-groups; the old names keep working for one version. - 1.2.1 (1 Oct 2026). Trending counts distinct outlets instead of URLs (a syndicated article under many URLs is one source). Readable endpoint names and descriptions.
- 1.2.0 (1 Oct 2026). The
/v2API: statement search, feed, time series, symbol and sector screeners, topics and themes with outlets and velocity, trending with newest statements, cursor paging, problem+json errors.
Guides
/v2/statements
Grounding an LLM in today's market news
Short, dated, sourced facts make better model context than whole articles. Build a prompt block about any ticker in a few lines of Python.
6 min read →/v2/feed
Never miss a statement: an alert stream with /v2/feed
A cursor feed in the order statements enter the API gives you every new fact exactly once, including late articles. Run it as an alert loop, and keep the log as your own point-in-time history.
6 min read →/v2/series
A news-attention and sentiment series for any ticker
How much is the news talking about a company, and in what tone? Hourly or daily, zero-filled, in one request, plus the normalisation that makes it comparable.
7 min read →/v2/symbols
Screening for unusual news attention
Which names is the news suddenly talking about? One request ranks every symbol by its change in mentions, outlets and tone, and one more does the same for sectors. Then filter it the way a quant should.
6 min read →/v2/topics
Detecting developing stories
A story is moving when more outlets pick it up, faster. Rank developments by velocity, check their breadth, and follow one from first report to peak.
7 min read →/v2/statements
Use Finance Pulse from Claude, Cursor or Python
One package gives you both a Python client for scripts and notebooks, and an MCP server so Claude, Cursor and other MCP clients can query the news directly.
5 min read →