edgar.tools · MCP server
Plug SEC data straight into your AI assistant over MCP — the connector standard behind Claude's and ChatGPT's integrations. Filings, financials, insider trades, and material events, answered in plain English where you already work, with every figure traced back to the filing it came from.
Works inClaude · ChatGPT · Cursor · custom agents
company_brief is free, and it's the widest single call in the set: profile, last annual figures, and market context in one answer. Ask for a company by name and your assistant gets this back.
Give me a brief on Nvidia — what do their latest filings say about how the business is doing?
edgar.tools returns
Fiscal year ended 2026-01-31. Nasdaq. Semiconductors & Related Devices.
Source: NVIDIA FY2026 10-K, filed 2026-02-25 · market data prior close, 2026-08-07 · sample captured 2026-08-08
Every number above is read out of the filing. The assistant is told to link the source rather than paste an accession number, so you can always click through.
Every 8-K in the trailing 90 days with its item codes and the price reaction that followed — the call the Track recipes are built on.
What has Nvidia filed 8-Ks about this quarter — and did the market care?
edgar.tools returns
The call flags the 5.02 officer change as the one high-impact item in the window — the rest is routine.
Each row links its SEC filing index · Price reactions are prior-close based, not real-time quotes · sample captured 2026-08-15
A recipe is a prompt that walks your assistant through a real analysis — which tools to call, in what order, what to do with the answers. All of them are free to take. What differs by tier is how far your assistant gets when it runs one. Browse the full library →
One filing, one ticker, fast. Get through a 10-K or an earnings release without reading all of it.
Decode the latest earnings release from the 8-K — reconcile the adjusted headline to GAAP, weigh growth quality and cash conversion, read the guidance delta.
Needs Pro to run
Calls
YoY diff of Item 1A across two 10-Ks. Buckets added / removed / expanded / tightened risks, cross-references 8-Ks, surfaces 2-3 load-bearing changes.
Runs partway
Calls
Six-step workflow for triaging a 10-K fast — business, numbers, real risks, capital allocation, post-10-K 8-Ks — grounded in primary-source MCP fetches.
Runs partway
Calls
Fills an existing comp sheet or monitoring book from as-filed XBRL — input cells only, an accession beside every figure, UNKNOWN where the filing is silent.
Runs partway
Calls
A watchlist on a cadence. Many tickers, run weekly, triaged down to what moved.
Weekly workflow for triaging a watchlist's 8-Ks: load-bearing vs context vs noise via SEC item codes and high-impact flags from material_events.
Needs Pro to run
Calls
Weekly Form 4/3/5 triage — strips mechanical noise (A/C/F/M/G + 10b5-1), ranks by role, flags clusters, cross-references the full 8-K list. Surfaces stale data.
Needs Pro to run
Calls
Long-form, framework-driven analysis across several filings and peers. The deepest thing the assistant will do unaided.
12-factor framework for grading compounders. Treats SBC as a real cost, mandates owner-earnings analysis, hard-gates broken per-share economics.
Runs partway
Calls
Builds the series of what a filer has promised but not recorded — leases not yet commenced, purchase commitments, VIEs — and files it as a cited dossier.
Runs partway
Calls
Assembles the person-keyed Form D cluster across every wave since 2001, tests whether money moved, sweeps counterparties, files a cited dossier.
Runs partway
Calls
Forensic bear-case framework — earnings quality, cash vs. accruals, solvency, disclosure red flags, insider selling — grounded in filings, never invented.
Runs partway
Calls
Ask your assistant for Nvidia's financial ratios today and it gets the margins. The rest of the answer comes back masked. Same question, same call — here's both sides of it.
How healthy is Nvidia's balance sheet? Give me the ratios.
Your assistant can only answer half the question, and it will say so.
Plus a health read — healthy — strong margins, liquid, low leverage — and every recipe above runs further.
Source: NVIDIA FY2026 10-K, filed 2026-02-25 · sample captured 2026-08-08
Cost side of the same decision: on our published benchmark, these tools answered the same SEC research questions with 10–20× fewer tokens than agentic web search, and scored higher for correctness. Methodology and per-run records →
Each row below is one thing your assistant can do — a tool it calls on its own, mid-conversation, when your question needs it. You never invoke them by name.
6 tools·100 calls/day
Find any company, browse its filings, get a one-call snapshot.
Find a company by ticker, CIK, or name prefix.
search_companiesFederated entity search across companies, funds, advisers, and persons.
search_entitiesOne-call company profile — identity and headline financials; Pro adds insider activity, recent events, and ownership.
company_briefBrowse a company's recent filings, filterable by form on Pro.
company_filingsA 13F manager's own portfolio — positions ranked by value with portfolio weight and quarter-over-quarter changes.
manager_holdingsFetch a prompt-framework recipe (FORGE, DCF, 10-K speedrun, etc.) to run inside your LLM with edgar.tools MCP grounding.
edgar_recipes$30/mo
17 tools·10,000 calls/day·7-day trial
Full financials, insider trades, institutional ownership, real-time material events, narrative-text disclosure + notes search, and fund + adviser intelligence.
Keyword search across XBRL TextBlock disclosures in a company's recent 10-K/10-Q.
disclosure_searchWhy a company is in (or out of) a saved universe — per-axis pass/fail with actual vs threshold.
explain_membershipAudit any screen verdict — which axis admitted or excluded a name, by how much.
List your saved universes and watchlists, or pull one's current roster and recent changes.
get_universe_membershipYour saved screens as agent-ready rosters — one call from universe to universe-scoped research.
Full-text search across SEC filings since 2001 — phrase, boolean, proximity, and wildcard operators; pass `interval` for a count-over-time trend instead of hits.
search_filingsFull RIA profile by CRD — AUM, executives, private funds, and compliance flags.
adviser_profilePull one section of a filing — MD&A, business overview, risk factors, an XBRL note, an 8-K item body, or the press-release exhibit (EX-99.1/99.2) behind it.
filing_sectionOne call, one section — pull the MD&A paragraph you actually need instead of fanning the whole 10-K through the agent.
Recent SEC filings from the real-time feed (last 48h), with form/CIK filtering on Pro.
live_filingsSide-by-side financial metrics across 2–10 companies — revenue, margins, growth, and 7 more.
compare_companies12 financial ratios from the latest 10-K — profitability, returns, liquidity, leverage — with citation.
financial_snapshotFull XBRL line items for income statement, balance sheet, or cash flow — TTM, annual, or quarterly.
financial_statementsMulti-year time series for revenue, earnings, and 9 other financial concepts — with YoY growth and CAGR.
financial_trendsForm 3/4/5 insider trades — buy/sell summary plus per-transaction detail.
insider_activityInstitutional (13F) holders for a stock, ranked by position size — with how fresh each holding is and quarter-over-quarter share moves.
institutional_ownership8-K events — per-company OR corpus-wide search by item code + date window.
material_eventsStructured, form-aware breakdown of any SEC filing — 8-Ks, 10-Ks, Form 4s, 13F-HRs, 424Bs, and NPORT-Ps — as context for analysis.
analyze_filingForm 4s get insider transactions; 8-Ks get material events; 10-Ks get financial summaries plus notes TOC; 13F-HRs get top holdings.
Save your synthesis as a durable, cited report with a private URL.
create_reportTurn an agent research session into a durable artifact — typed blocks, every claim cited.
One-call filing reader: open a filing for its digest + retrievable-part manifest, or fetch a named part — the open→drill loop of analyze_filing + filing_section in a single tool.
filing_detailsOpen a filing, read the sized part manifest, drill into exactly the sections you need — one tool for the whole loop.
$109/mo
3 tools·50,000 calls/day·7-day trial
Whole-filing AI analysis, plus aggregation bundles that answer across a position, a peer set, or an entire portfolio in one call — and the slash-command workflows that chain them for you.
Screen the profiled-company universe with a routed thesis plan — semantic, financial, capability, and similarity clauses in one call.
screen_companiesThe Companies Hub screening engine as a tool — your model plans, our engines answer with cited evidence.
One-call sales dossier for an account — identity, financials, recent 8-Ks, leadership, insider signals, and peers, with hop-able CIKs and accessions.
account_dossierReplaces the 5–8 call fan-out per account with one call — the anchor of a sales research session.
Everything one person has filed — insider roles, 13D/G activist stakes, ADV control roles, and trades — in one call.
person_intelThe enterprise/compliance person graph — auditor independence, D&O background, and activist tracking from one call.
Need bulk data instead of live tool calls? Browse the dataset catalog →
Slash commands
These ship with the connection: type one in your client, fill in the arguments, and Claude runs the tool calls and writes up the answer — accession-cited and structured. Two different things share the word "recipe", so to be precise: these are slash commands, defined by the server and listed in your client's own menu. The recipe library is prompts you copy, adapt, and paste yourself. The slash commands are the ones that keep showing up in your client without you doing anything — still there the next time you open it, for the next name on your list.
/edgar:earnings_postmortemAnalystPost-mortem an earnings cycle: beats/misses, guide changes, Q&A red flags.
Argumentsticker · since?
/edgar:earnings_postmortem ticker=NFLX since=2026-02-11Three minutes in your AI client of choice. Same OAuth flow, same tool surface, same SEC data.
Settings → Integrations → add the edgar.tools URL. OAuth completes in your browser.
// claude_desktop_config.json
{
"mcpServers": {
"edgar-tools": {
"url": "https://app.edgar.tools/mcp"
}
}
}Add the connector once in your Claude settings and it carries into Excel — Claude writes financials, comps, and holdings into your workbook as cells. Claude for Excel guide.
# claude.ai → Settings → Connectors
Name: edgar.tools
URL: https://app.edgar.tools/mcp
# in Excel: + → Connectors → edgar.toolsConnectors → Add custom connector. Available on Plus and above; Enterprise and Team workspaces supported.
# Add custom connector
URL: https://app.edgar.tools/mcp
Auth: OAuth 2.1 (browser flow)Standard MCP server endpoint with bearer token. Works with the Anthropic SDK, OpenAI Agents SDK, LangGraph, and any MCP client.
# Test from your terminal first
curl -X POST https://app.edgar.tools/mcp \
-H "Authorization: Bearer $EDGAR_TOKEN" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'Quick answers to the questions analysts and developers ask before they connect.
Ask it the question you would otherwise open a filing to answer. Each question below maps to one tool your assistant calls on its own:
Are insiders buying or selling? → insider_activity reads the Form 4s (an insider's own trade report).
What did this company just announce? → material_events reads the 8-K (a same-day disclosure of something material).
What are the numbers, without reading the annual report? → financial_snapshot for the headline figures, financial_statements and financial_trends for the statements and the multi-year shape.
Who owns it, and what does a fund hold? → institutional_ownership and manager_holdings read the 13F (a fund's quarterly holdings report).
Where does the language actually say that? → disclosure_search runs phrase-level search across narrative disclosures and financial-statement notes; pass context: true to read every match as one block.
What does that whole filing actually say? → analyze_filing opens any filing — an 8-K, a 10-K, a Form 4 — into a structured, form-aware breakdown, and filing_section pulls the exact passage behind it.
On Analyst, the assistant stops working one company at a time: screen_companies builds a universe from a plan, get_universe_membership and explain_membership read it back with per-row evidence, person_intel follows one person across every filing they touch, and create_report writes the result up.