Filing Q&A
Let users ask questions across 10-Ks, 10-Qs, 20-Fs, 6-Ks, and 8-Ks with source-linked answers.
AlphaCreek gives your agents a hosted MCP connection to structured, citation-ready SEC filings — so your team can focus on product logic, not ingestion pipelines.
Connect from Python, JavaScript, or any MCP-compatible client. Skip EDGAR crawlers, chunking experiments, and citation plumbing.
# 1. List filings for a ticker
await mcp.call("list_filings", {ticker: "NVDA"})
# 2. Navigate the filing map
await mcp.call("get_filing_toc", {artifact_document_id: "…"})
# 3. Read targeted nodes with citations
await mcp.call("read_node_content", {node_ids: ["item-1a-risk-factors"]})
→ CITATION_URL per NODE_ID for verifiable answers
Most teams power financial AI by stitching together complex stacks from scratch. AlphaCreek replaces brittle pipelines and months of engineering with a single MCP connection.
Months of engineering. Brittle infrastructure.
One MCP connection. Production-ready in days.
Use AlphaCreek when your AI product needs filing-backed answers, source-linked context, and a retrieval path your users can verify.
Let users ask questions across 10-Ks, 10-Qs, 20-Fs, 6-Ks, and 8-Ks with source-linked answers.
Retrieve and compare Item 1A disclosures across periods.
Pull management commentary, operating drivers, liquidity discussion, and margin explanations.
Retrieve original filing language behind segment, product, and geographic performance.
Give agents traceable context for business, legal, accounting, and risk disclosures.
Bring SEC filings into Claude, Perplexity, Cursor, Claude Code, or another MCP-compatible workflow.
Snippets are illustrative only. Follow the link on the code card for runnable examples and the MCP reference.
1# Illustrative — Claude + AlphaCreek MCP (Streamable HTTP).
2# Runnable script: project_ff/alphacreek_mcp/examples/python_anthropic.py — docs: /docs#python-anthropic-example
3
4import asyncio
5import functools
6import mcp.types as T
7from anthropic import Anthropic
8from mcp import ClientSession
9from mcp.client.streamable_http import streamablehttp_client
10
11MCP_URL = "https://mcp.alphacreek.ai/mcp"
12HEADERS = {"Authorization": "Bearer <ALPHACREEK_API_KEY>"}
13QUESTION = "What are NVIDIA's biggest risks in the latest 10-K?"
14
15async def ask_with_filing_tools() -> None:
16 ac = Anthropic(api_key="<ANTHROPIC_API_KEY>")
17 async with streamablehttp_client(MCP_URL, headers=HEADERS, timeout=120, sse_read_timeout=300, terminate_on_close=True) as (read, write, _):
18 async with ClientSession(read, write, client_info=T.Implementation(name="demo", version="0")) as session:
19 await session.initialize()
20 listed = await session.list_tools()
21 atools = [
22 {
23 "name": t.name,
24 "description": ((t.description or "").strip() or "(MCP tool)"),
25 "input_schema": t.inputSchema if isinstance(t.inputSchema, dict) else {"type": "object", "properties": {}},
26 }
27 for t in listed.tools
28 ]
29 messages = [{"role": "user", "content": QUESTION}]
30 call = functools.partial(
31 ac.messages.create,
32 model="claude-sonnet-4-6",
33 max_tokens=8192,
34 system="Answer using only AlphaCreek filing tools; ground claims in tool output.",
35 messages=messages,
36 tools=atools,
37 tool_choice={"type": "auto"},
38 )
39 resp = await asyncio.to_thread(call)
40 # If stop_reason is tool_use: await session.call_tool(...), append assistant + user tool_result
41 # blocks to messages, then messages.create again until end_turn (see /docs).
42We transforms every filing into high-signal, LLM-ready context so your AI can reason over real financial data with precision.
We generate a table of contents for every filing we ingest. This allows AI agents to understand the filing structure and then navigate to the most relevant sections efficiently, instead of reading the entire document or searching naively.
The smart navigation allows agents to consume only what's important for the query at hand. The result: dramatically lower token consumption, faster responses, and more model headroom for the reasoning that actually matters.
Each filing is broken down into referable logical chunks. This enables agents to answer questions with precision and cite the exact parts of the filing that were used to answer the question.
Start with self-serve docs and an API key. If you are planning a production integration, talk to us so we can help with rollout, limits, and the right retrieval workflow.
Create a free AlphaCreek account and generate an API key from your dashboard for programmatic MCP access.
Follow tool schemas, auth flows, and runnable Python or JavaScript examples for OpenAI, Anthropic, and Gemini.
Use hosted MCP with OAuth or API keys, align on expected usage, and give your users citation URLs they can verify.