edgar.tools MCP · Prompt library

MCP prompt recipes for SEC-filing analysis

MCP prompt recipes are copy-paste LLM prompts that call edgar.tools MCP tools to run real SEC-filing analysis. FORGE for compounding-quality scorecards, 10-K speedrun for load-bearing facts, 8-K triage for event-day decisions, Form 4 reader for insider patterns, risk-factor diff for what changed year over year. Every number traces back to an EDGAR filing — not the model's training data.

5 recipes live·Free to copy·Open in any LLM

01/Section

Read

Workflows for getting through a single filing fast — 10-K, earnings release, risk-factor diff. One ticker at a time.

2 recipes

02/Section

Track

Workflows for monitoring a watchlist or pipeline — 8-K triage, insider activity, material event triagers. Many tickers, recurring cadence.

2 recipes

03/Section

Research

Deeper, longer analyses with framework- or super-investor-style output — FORGE, DCF, Piotroski, Buffett-style, Klarman-style. Grounded in real filings via peer comparison and multi-tool composition.

1 recipe
Featured recipe
F.O.R.G.E. Framework — Compounding Engine Analysis

12-factor framework for grading compounders. Treats SBC as a real cost, mandates owner-earnings analysis, hard-gates broken per-share economics.

financial_trendsfinancial_snapshotfiling_sectionpeer_facts+2
Try it onASANOpen recipe →

Frequently asked

Quick answers to what analysts and developers ask before they paste a recipe into their AI client.

What is an MCP prompt recipe?
A prompt recipe is a copy-paste LLM instruction that orchestrates edgar.tools MCP tools into a structured analysis. You paste it into Claude, ChatGPT, Cursor, or any MCP-compatible client, give it a ticker, and the model walks the recipe — pulling 10-K narrative, financial statements, insider activity, or whatever the recipe specifies — then returns a formatted answer. The structure is fixed; the data is live.
Do I need a paid plan to use these recipes?
The recipes themselves are free to copy and read. Running them requires the MCP tools they reference, which are tier-gated. Most recipes work on the Pro tier ($24.99/mo); FORGE and Analyst-tier recipes need analyze_filing or the aggregation tools, which are on Analyst ($79.99/mo). Each recipe page lists the exact MCP tools it depends on, so you can check tier requirements before running.
Can I use these recipes in ChatGPT, Cursor, or Claude Desktop?
Yes. Recipes are plain LLM prompts designed to work in any MCP-compatible client — Claude Desktop, Claude.ai, ChatGPT Connectors, Cursor, Windsurf, or a custom agent built on the Anthropic SDK / OpenAI Agents SDK / LangGraph. Once edgar.tools is added as an MCP server, the same recipe text works across clients with no per-platform adaptation.
How is a recipe different from a regular prompt template?
A regular prompt is static text. A recipe orchestrates live tool calls — it tells the model to fetch a specific filing, extract typed facts, compute ratios, and synthesize the result. The output is grounded in current SEC data instead of the model's training knowledge. Recipes also enforce shape: every FORGE run returns the same scorecard sections; every 10-K speedrun produces the same load-bearing-facts breakdown.
Where does the data come from?
Every fact a recipe surfaces is pulled from a SEC EDGAR filing at call time, via the edgar.tools MCP tools. Financial statements come from XBRL; narrative text from 10-K and 10-Q sections; insider trades from Form 4; material events from 8-K item codes; fund holdings from NPORT-P. No third-party financial-data vendor, no training-data fallback, no unverifiable source.