arkleon

arkleon is the free client for Arkleon's point-in-time SEC fundamentals: filing-dated facts straight from EDGAR with no key, and an optional API client for the certified corpus (16,811 companies, 426,003 filings, 181,350,662 facts) with replayable as_of queries.

The free EDGAR core requires no account and no API key. The optional paid /v1 client and the corpus-backed point-in-time features are the only parts that need a key.

Install

pip install arkleon

Quickstart: free, no API key

The EDGAR helpers work on the first line of code with no Arkleon account and no key. The only string they require is an SEC fair-access User-Agent, which you write yourself.

from arkleon.edgar import EdgarClient

# SEC fair-access User-Agent: a self-declared contact string, NOT an Arkleon key.
edgar = EdgarClient(user_agent="Acme Research contact@acme.com")

# Fetch a company's 10-K filings by CIK (Apple = 320193):
for filing in edgar.filings(cik=320193, form="10-K"):
    print(filing.filed, filing.accession, filing.form)

# Parse as-reported numeric facts, filtered best-effort to a date:
assets = edgar.concept(cik=320193, tag="Assets", taxonomy="us-gaap")
for fact in assets.as_of("2020-01-01"):        # keep only filed <= 2020-01-01
    print(fact.period_end, fact.value, fact.unit, fact.filed)

No account, no key, no Arkleon network. Live-EDGAR, best-effort point-in-time.

Quickstart: optional paid /v1 client (certified point-in-time)

The free /v1 tier allows 500 requests / day.

A concept_alias parameter (plain-language concept names over raw us-gaap tags) is reserved and available after /v1 ships it.

# pip install "arkleon[api]"
from arkleon.api import DataClient

client = DataClient(api_key="ak_...")          # or ARKLEON_API_KEY env
page = client.facts(cik=320193, tag="Assets", as_of="2020-01-01")
for fact in page.data:
    print(fact["period_end"], fact["value"], fact.get("source_url"))

Certified corpus. as_of required, filters on filing date, identical query + as_of returns identical data permanently.

The free/paid boundary

The arkleon.edgar helpers never require a key and never contact Arkleon: their only remote hosts are data.sec.gov and www.sec.gov. Only the arkleon[api] client and the corpus-backed point-in-time features call Arkleon, and they refuse to operate without an ak_-prefixed key. Installing arkleon[api] does not by itself activate the paid path; a key must still be supplied at runtime.

MCP server

Install the server extra and launch the stdio server with the console script:

pip install "arkleon[mcp]"
arkleon-mcp

An agent host launches it the same way. Set ARKLEON_API_KEY in the launch environment to unlock the paid tools:

{
  "mcpServers": {
    "arkleon": {
      "command": "arkleon-mcp",
      "env": { "ARKLEON_API_KEY": "ak_..." }
    }
  }
}

The free EDGAR tools register unconditionally; the paid tools register only when an ak_ key is present. An agent with no key sees exactly the free set.

Tool	Backed by	Key required?
edgar_list_filings	free EDGAR core	No
edgar_company_facts	free EDGAR core	No
edgar_concept	free EDGAR core	No
edgar_facts_as_of (best-effort, live EDGAR)	free EDGAR core	No
edgar_resolve_cik (current-snapshot, non-point-in-time)	free EDGAR core	No
pit_facts (certified point-in-time)	paid /v1 client	Yes
pit_filings	paid /v1 client	Yes
resolve_company	paid /v1 client	Yes
pit_revision_history	paid /v1 client	Yes

Registry: The server is described by server.json at the repository root for the official MCP registry under the name io.github.jushuea/arkleon. It launches over stdio as arkleon-mcp; the free EDGAR tools register unconditionally and the paid tools only when an ak_ key is present. Install with pip install "arkleon[mcp]".

LangChain and LlamaIndex

Both integrations wrap the free EdgarClient and need no Arkleon key. The SEC requires a descriptive User-Agent, so pass user_agent, pass client=EdgarClient(...), or set SEC_EDGAR_USER_AGENT.

pip install "arkleon[langchain]"
pip install "arkleon[llamaindex]"

ArkleonEdgarFactsTool is a LangChain tool named arkleon_edgar_facts. Arguments are cik, tag, a required as_of in YYYY-MM-DD form, and taxonomy, which defaults to us-gaap.

from arkleon.integrations.langchain import ArkleonEdgarFactsTool

tool = ArkleonEdgarFactsTool(user_agent="Name email@example.com")
result = tool.invoke({"cik": 320193, "tag": "Assets", "as_of": "2020-01-01"})
print(result["facts"][0]["value"])

It returns a dict with cik, tag, taxonomy, as_of, warning and facts. Each fact has period_start, period_end, value, unit, form, filed, accession, instantaneous and the other Fact fields.

ArkleonEdgarFactsReader is a LlamaIndex reader. load_data returns 1 Document per fact, with the fact fields plus cik, tag, taxonomy, as_of and warning in its metadata.

from arkleon.integrations.llamaindex import ArkleonEdgarFactsReader

reader = ArkleonEdgarFactsReader(user_agent="Name email@example.com")
docs = reader.load_data(cik=320193, tag="Assets", as_of="2020-01-01")
print(docs[0].metadata["filed"])

as_of keeps only facts filed on or before that date, which is best-effort over live EDGAR and not certified, and every result carries that warning. Neither integration wraps the paid /v1 DataClient.

What it returns, and what it does not

Returns: as-reported numeric XBRL facts, filing and company metadata, and the lineage linking each figure to its source filing.

Does not return: market prices, analyst estimates, narrative disclosure (MD&A, risk factors, footnotes), or derived analytics, ratios, scores, or signals. Neither layer serves these. It is a research corpus, not a real-time filing feed: the paid corpus refreshes on the SEC's quarterly cadence.

Point-in-time, in two sentences

as_of filters on the filing date (filed <= as_of), never on the period a fact describes, so a query sees only values that were already public on that date. Because a fact from a later filing never appears, a historical query cannot observe information that did not yet exist, which is what keeps a study free of look-ahead bias.

The certified corpus (paid)

The paid /v1 client serves the certified corpus. Per the /v1 API contract at https://arkleon.com/docs, which is the sole authority for what any surface may state, the claimable figures are:

Quantity	Figure	Kind
Filings	426,003	EXACT
Facts	181,350,662	EXACT
Distinct companies (CIK)	16,811	EXACT
Quarters spanned	69 (2009q1–2026q1)	exact
Quarters populated	68	exact

The EXACT/exact label travels with each number and is not presentational. These figures are keyed to the date each filing became public, include values that later filings revised, and trace to their source filing. The earliest filed date in the corpus is 2009-04-15, so a query with an as_of earlier than that date correctly returns empty. Treat the contract's Appendix A, not this table, as the authority: a later corpus revision updates the contract, and any figure that disagrees with it is stale.

Identifiers

CIK is the addressable identifier across both layers. In the free layer, resolve_cik(ticker) maps a symbol to a CIK from SEC's current ticker snapshot; it is a convenience, not a reproducible identity resolver, because it has no ticker history, a symbol is reassigned over time, and one CIK carries several symbols at once (GOOG and GOOGL, BRK.A and BRK.B). The /v1 API does not serve the ticker filter today: a ticker request returns a distinct "not served in this release" error rather than silently resolving through the free snapshot. Address facts by CIK.

Install matrix

Install	Adds
arkleon	The free EDGAR fetch/parse core. No key, no Arkleon network.
arkleon[api]	The optional paid /v1 DataClient. Requires an ak_ key at runtime.
arkleon[mcp]	The built-in MCP server and the arkleon-mcp console script.
arkleon[pandas]	Optional pandas DataFrame adapters for the free core. Never required.
arkleon[langchain]	The ArkleonEdgarFactsTool LangChain tool and its langchain-core dependency.
arkleon[llamaindex]	The ArkleonEdgarFactsReader LlamaIndex reader and its llama-index-core dependency.
arkleon[all]	Convenience union of api, mcp, and pandas.

Base plus [pandas] stays fully free and credential-free. [api] and the paid tools in [mcp] are the only parts that consume an ak_ key.

License

MIT. See LICENSE.

Links

• /v1 API contract: https://arkleon.com/docs (the authority for the corpus figures above).
• Issues and contributing: https://github.com/jushuea/arkleon-python
• SEC fair-access policy: https://www.sec.gov/os/accessing-edgar-data