About EDGAR Filings
EDGAR Filings is an MCP server for people using IDE chats to retrieve specific SEC filing data without writing edgartools code. It covers four targeted tasks and returns filing references for verification.
Key features
- 01Latest 10-K/10-Q trading symbols
- 02Dimensioned XBRL segment revenue
- 03BDC non-accrual data and method
- 04Form 4 transactions and open-market flag
Use cases
- 01Check a company’s reported trading symbols
- 02Compare product or geographic segment revenue
- 03Review a BDC’s non-accrual figures
- 04Inspect recent insider transaction lines
MCP support
Exposes four SEC filing tools to MCP clients, including Cursor and Claude. Results include filing identifiers and EDGAR index URLs for checking the source.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
uvx edgar-filings-mcpEDGAR_IDENTITYSEC fair-access User-Agent: 'Name [email protected]'Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
4Saved tool list · updated 2w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| form | string | No | |
| accession | any | No | |
| ticker_or_cik | string | Yes |
Details
From the project READMEEDGAR Filings MCP
Cursor / Claude tools that return SEC filing numbers, not web-search guesses.
Who this is for: people already in an IDE chat who need a ticker’s 10-K/10-Q symbols, segment revenue, BDC non-accrual, or Form 4 lines without writing edgartools glue.
Pain it solves: models invent segment mix, insider trades, and BDC credit quality. These tools return accession, concept, period, open_market, non-accrual method, and the EDGAR index URL so you can check the filing.
What it is not: a research product, a document reader, or a substitute for reading the 10-K. It wraps edgartools for four jobs only. China PE, humanoid robots, and unlisted credit CVs are out of scope — those filings are not on EDGAR.
Install
If uv is already on PATH, paste this into ~/.cursor/mcp.json (Windows: %USERPROFILE%\.cursor\mcp.json). Use a real name and email (SEC FAQ). Same snippet is in examples/cursor.mcp.json and examples/claude.mcp.json.
{
"mcpServers": {
"edgar-filings": {
"command": "uvx",
"args": ["edgar-filings-mcp"],
"env": {
"EDGAR_IDENTITY": "Your Name [email protected]"
}
}
}
}
PyPI package edgar-filings-mcp is live. Registry name is io.github.Dxfory/edgar-mcp. Skip Smithery hosted. Pin mcp>=1.9,<2 — MCP 2.x renamed FastMCP; the package already pins that range.
The first uvx launch downloads edgartools (pandas / pyarrow). If the client looks stuck, run uvx edgar-filings-mcp once in a terminal so uv can cache the wheels, then restart the MCP server. After that, initialize is a couple of seconds.
No uv yet:
macOS / Linux:
curl -LsSf https://astral.sh/uv/install.sh | sh
Windows (PowerShell):
irm https://astral.sh/uv/install.ps1 | iex
From git instead of PyPI:
uvx --from git+https://github.com/Dxfory/edgar-mcp.git edgar-filings-mcp
Clone (no uv)
git clone https://github.com/Dxfory/edgar-mcp.git
cd edgar-mcp
python -m venv .venv
macOS / Linux:
.venv/bin/python -m pip install -e .
Windows:
.\.venv\Scripts\python -m pip install -e .
Point the client at that interpreter with args ["-m", "edgar_mcp"].
If PyPI SSL fails (common with a local HTTPS proxy), use a mirror:
python -m pip install -e . -i https://pypi.tuna.tsinghua.edu.cn/simple
Tools
| Tool | Returns |
|---|---|
get_trading_symbols | Every dei:TradingSymbol on the latest 10-K or 10-Q, plus the legacy entity_info scalar |
get_segment_revenue | Dimensioned XBRL revenue (product / business / geographic axes) |
get_bdc_nonaccrual | BDC non-accrual rate, fair value, named investments, and extraction method |
get_form4 | Newest Form 4 summaries, transaction lines, open_market, and code_counts |
form on get_trading_symbols, get_segment_revenue, and get_bdc_nonaccrual is 10-K (default) or 10-Q. get_bdc_nonaccrual only accepts SEC BDCs (814- filers) such as ARCC. There is no fifth tool.
Hot-theme footguns this server will / will not answer
| Theme | Agent invents | Tool | Stop |
|---|---|---|---|
| AI infrastructure | NVIDIA / hyperscaler “AI mix” | get_segment_revenue on NVDA (Data Center is tagged). AMZN/MSFT capex is not AI-only | Do not add a fake AI-capex tool |
| Private credit / BDC | Non-accrual, NAV as credit quality, PIK as current | get_bdc_nonaccrual | Non-accrual ≠ Fitch default rate; PIK can still be accrual |
| GP-led continuation vehicles | Deal price and “premium to par” | None | Private secondaries are not EDGAR |
| China PE / 具身智能 | Round sizes and factory hours | None | SSE/HKEX, not EDGAR |
Footguns the tools already warn about
entity_info.tickeris last-wins on repeatedTradingSymbolfacts. Dual-class and preferred tickers can replace the common symbol.- Segment mix is not in
get_financials(). Some statement “DETAILED” views drop reportable-segment lines; this server queries dimensioned facts instead. - Form 4
A/M/Fare grants, option exercises, and tax withholding —open_marketis false. - BDC
extraction_method=noneor a zero rate plus extractor warnings is a parse gap, not proof the book is clean. - Latest
10-Kcan be a10-K/A. The tools prefer the original form so Schedule-of-Investments footnotes are not dropped.
Run without Cursor
python -m edgar_mcp
stdio only. Do not print to stdout.
python tests/run_offline.py
python scripts/smoke_stdio.py
python scripts/pressure.py
smoke_stdio.py only checks initialize + four tool names (no EDGAR). After uvx is on PATH:
python scripts/smoke_stdio.py -- uvx edgar-filings-mcp
pressure.py hits live EDGAR and needs EDGAR_IDENTITY.
License
MIT. Filing data is from the SEC EDGAR system; this project is not affiliated with the SEC.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.