Turn your AI into
an equity analyst.
Look up any BSE-listed company — FilingForge gathers its official filings, organizes them, and turns them into clean text your AI can finally read. Local-first. Free. Open-source.
No account · your data never leaves your computer.
Institutions pay a fortune for clean, structured filings. Hand your AI the same primary sources — every official BSE filing, in a format it can actually use — and it does expert-grade equity research for you. On your own machine.
A look inside
From a company lookup to an AI-ready library — the whole flow, in a few clicks.
How it works
A company's filings are dozens of PDFs scattered across BSE — and your AI can't read any of them cleanly. FilingForge fixes the whole chain, end to end.
Pull every filing
Find any BSE-listed company and download all its official filings — annual reports, results, concalls, investor presentations — not just the one you remembered.
Organize & rename
Sorted into clean year-wise folders by category, with cryptic BSE filenames renamed to readable ones. A library you can navigate by hand.
Convert to Markdown
Every PDF becomes structured Markdown your AI can actually read — tables intact, numbers aligned, layout noise gone. (No OCR or GPU; scanned PDFs are kept and flagged.)
Index for your AI
A per-company INDEX.md and a master INDEX.md across all of them — point your AI at one folder and it reads the entire library, with sources.
What step 03 actually does — the same filing, before and after:
%PDF-1.4 … /F2 9 0 R Revenue 1,2 4 8.3 2 ,1 09.7 (cid:32)(cid:71) [table spans pages] running headers, footers, page nums …34,000 tokens of layout noise
## Revenue (₹ cr) | Segment | FY24 | FY23 | |---------|-----:|-----:| | Retail | 1,248| 1,109| | Wholesale| 860 | 740| *Source: Annual Report FY24, p.42*
What's new
FilingForge keeps getting better at the hard part: finding the real document and handing your AI clean text.
Annual reports back to 1997
FilingForge reads BSE's separate annual-report archive too, so you get reports from 1997 onward — including companies that have since been delisted. Refresh a company and the missing years fill in; nothing you already have is downloaded again.
Works with Claude
A built-in MCP server lets Claude Desktop and Claude Code browse your companies, search filings and read the clean Markdown themselves — no copying files around.
Follows cover-letter links
When a BSE attachment is just a notice pointing elsewhere, FilingForge safely finds the real annual report, concall, presentation or results when there is one clear official PDF.
Reliable multi-year history
Long date ranges are fetched year by year, so a library no longer ends up with only annual reports. A failed year is reported, never silently skipped.
Coverage: any BSE-listed company. Most NSE-listed companies are also listed on BSE, so you'll find them here too.
Use it with Claude
Point Claude Desktop or Claude Code at your library and it works the filings itself. Your files stay on your computer; the server can only read the folder you give it.
pip install "filingforge-engine[mcp]" filingforge-mcp --root ~/FilingForge
Then ask Claude things like “Pull the last five annual reports for this company and summarise how the business mix changed.” It reads the clean Markdown and cites the source file. Setup details on GitHub.
Skills
Skills are optional prompt-packs that turn your clean library into finished work — run in your Claude desktop app or Codex. FilingForge never calls an LLM; you bring your own.
- Pick a Skill and a company. FilingForge writes a ready-to-run prompt with that company's exact file paths already filled in, copied to your clipboard.
- Paste it into your AI. It reads the real filings and hands back the finished, fully-cited result — a polished report you can share.
Make your own. A Skill is just a Markdown file — write one or import one, and it runs like any built-in.
Two skills, more over time. Business Model Brief is free; Concall Decoder — every earnings call decoded into a guidance hit-rate and a read on how far to trust management — is ₹3,000, one-time. Buy it in-app: pay in your browser, it unlocks automatically.
