AI Coding Tools Integration🔗

AI coding assistants generate better code when they have project-specific context. Each tool reads a different configuration file at session start. EDA provides ready-to-use files for both, generated directly from the extractor source code.

CLAUDE.md

Context file for Claude Code (Anthropic). Drop it in your repo root — auto-loaded at session start.

Download CLAUDE.md
AGENTS.md

Context file for Codex, Cursor, Aider and any tool supporting the AGENTS.md standard.s

Download AGENTS.md
llms.txt

Full AI context bundle hosted on the docs site. Use as a URL reference or Claude connector.

Overview🔗

Tool	Config file	Auto-loaded	Also read by
Claude Code (Anthropic)	`CLAUDE.md`	✅ Yes	Claude Code only
Codex CLI / Agent (OpenAI)	`AGENTS.md`	✅ Yes	Cursor · Aider · Builder.io

Two files, not one

Neither tool reads the other's configuration file. If your team uses both Claude Code and Codex, place both CLAUDE.md and AGENTS.md in your repo root. EDA ships both — they share the same core content.

What the Files Contain🔗

Both files share the same core structure. Content is generated from source — never manually edited, always in sync with the codebase.

CLAUDE.md / AGENTS.md — structure

# EDA Extractors — AI Context

## Architecture
# BaseExtractor → EDAuthenticator, loguru, ThreadPoolExecutor
# All extractors follow: .authenticate() → .extract(aoi, date_range, ...)

## Extractor Index
- CoverageExtractor    # Crop area coverage % per field
- EmergenceExtractor   # Planting emergence timing
- VegetationExtractor  # NDVI / vegetation health
- DiseaseRiskExtractor # Daily disease risk scores
- TillageExtractor     # Tillage practice detection
  ... all extractors with docs + notebook links

## Common Patterns
# aoi: GeoJSON polygon or [xmin, ymin, xmax, ymax]
# date_range: ("2024-01-01", "2024-12-31") — always ISO strings
# Output: GeoDataFrame unless as_dataframe=False

## Per-Extractor Sections
# Parameters, example snippet, output schema, raises

## Error Reference
# EDAuthError · ExtractionError · GeometryError · ...

Tool Compatibility🔗

Tool	`CLAUDE.md`	`AGENTS.md`	MCP connector	`llms.txt` via URL
Claude Code	✅ Native	✗	✅ Native	✅ Via `CLAUDE.md` ref
Claude.ai (web)	✗	✗	✅ Settings → Connectors	✅ Paste URL in chat
Codex CLI / Agent	✗	✅ Native	⚡ stdio only	✅ Via `AGENTS.md` ref
Cursor	✗	✅ Native	✅	✅ Paste URL
Aider	✗	✅ Native	✗	⚡ Manual
GitHub Copilot	✗	✗	✗	⚡ Manual paste

Setup in 3 Steps🔗

1. Download and place the file(s) in your repo root

Drop CLAUDE.md for Claude Code, AGENTS.md for Codex/Cursor, or both if your team uses multiple tools. No other configuration is required — each tool picks up its file automatically at session start.

2. Start your AI session as normal

Run claude or codex in your project directory. The tool loads the context file before your first prompt. Verify it was read by asking:

"What extractors does EDA provide?"

3. Optionally add the Claude connector for persistent access

In Claude.ai, go to Settings → Connectors → Add custom connector and enter https://docs.eda.com/mcp. This gives every Claude conversation live access to EDA extractor knowledge without needing any file in the repo.

Keeping Files Current🔗

Both files are generated by scripts/generate_ai_context.py on every push that touches the extractors or documentation. You never need to update them manually.

.github/workflows/release.yml

on:
  push:
    paths:
      - 'src/extractors/**'
      - 'docs/**'
      - 'notebooks/**'

jobs:
  update-ai-context:
    steps:
      - run: python scripts/generate_ai_context.py
        # Writes CLAUDE.md, AGENTS.md, llms.txt, and report.md
      - run: git commit -am "chore: update AI context" && git push

Documentation gap tracking

The same script generates ai_context/report.md — a per-extractor completeness report showing which classes are missing descriptions, examples, or output schemas. View documentation report →

Where Files Live🔗

Depending on your setup, files can live in the extractor repo, in the client's project repo, or just as a hosted URL — they don't all need to be in the same place.

Artifact	Lives in	Maintained by	Updated
`CLAUDE.md` / `AGENTS.md`	Client's repo	Client (one-time copy)	On major EDA API changes
`llms.txt` bundle	EDA docs site	EDA CI pipeline	Automatically on every build
MCP connector	Claude settings	Each developer, once	Never — always live

Info

The bundle is the single source of truth. The config files are lightweight wrappers that reference it. Once the MCP connector is added to Claude settings, CLAUDE.md becomes optional — Claude already has full context before the repo is opened.