Skip to main content

v0.4.1

Released 2026-05-01 · GitHub Release

Overview​

  • Improve AI integration
    • add MCP server at /mcp
    • publish llms.txt / llms-full.txt
    • support markdown content negotiation on docs
    • add a per-page Copy page / Open in chat widget
  • Make BalcanTransform.max_value optional (default 1.0); recommended pattern is to set only min_value

Improved AI integration​

MCP server​

The API now exposes itself as a Model Context Protocol server at:

https://epyscenario-api.isi.it/mcp

Setup snippets for each client are in the new AI Integration docs page. The MCP layer can be disabled per deployment with EPYDEMIX_MCP_ENABLED=false. An MCP Server Card (SEP-1649) is also published at /.well-known/mcp/server-card.json for agent discovery.

llms.txt and llms-full.txt​

The docs site now publishes the llmstxt.org resources:

  • /llms.txt: short index of every doc page (~30 lines)
  • /llms-full.txt: every doc page concatenated as a single markdown file (~37 KB)

Both are advertised in /.well-known/api-catalog. Generated from the source MDX files by scripts/generate_llms_txt.py on every Vercel deploy.

Markdown content negotiation​

When the docs site receives Accept: text/markdown, it returns the markdown source instead of the rendered HTML page:

# HTML (default for browsers)
curl https://epydemix-webapi.vercel.app/docs/intro

# Markdown (for AI agents)
curl -H 'Accept: text/markdown' https://epydemix-webapi.vercel.app/docs/intro

Vary: Accept is set so caches keep both representations separate.

Copy page / Open in chat widget​

Every doc page now has a combo widget at the top right with two click zones:

  • Copy page: copies the page as markdown to the clipboard
  • Open in chat: dropdown for Open Markdown / Open in Claude / Open in ChatGPT, each opening a new tab with the page URL prefilled in the prompt

Driven by per-page markdown mirrors at /docs/<path>.md that the same script writes alongside llms-full.txt.

Simulations​

POST /v1/simulations

BalcanTransform max_value is now optional​

max_value defaults to 1.0. The recommended pattern is to leave it unset and only configure min_value to express the seasonal floor as a fraction of the baseline:

{
  "target_parameter": "transmission_rate",
  "method": "balcan",
  "max_date": "2024-01-15",
  "min_date": "2024-07-15",
  "min_value": 0.1
}

Existing requests that pass max_value explicitly continue to work unchanged. The multiplier shape depends only on the ratio min_value / max_value, so (max_value=0.4, min_value=0.1) produces exactly the same dynamics as (max_value=1, min_value=0.25).

Documentation​

  • New: "AI Integration" page covering chat-style usage and MCP setup for Claude Desktop, Claude Code, Cursor, Codex, and Claude.ai web
  • Updated: Seasonality docs reflect the new optional max_value and clarify the multiplier semantics