Prompt Edit for Agents

Every page of the prompt engineering reference is available as clean Markdown, without scraping HTML. This page documents how to fetch it. Everything here is read-only, unauthenticated and safe to cache.

Content negotiation

Send Accept: text/markdown to any page URL and you get Markdown instead of HTML. The response carries Vary: Accept, a Content-Locationheader naming the .md URL, and x-markdown-tokens with an approximate token count.

curl -H 'Accept: text/markdown' \
  https://www.promptedit.app/prompt-framework/costar

If your client cannot set request headers, append .md to the path instead. The two are equivalent:

curl https://www.promptedit.app/prompt-framework/costar.md

Each Markdown document opens with YAML frontmatter carrying the page title, its meta description and the canonical source URL, so a retrieved document can always be cited back to the page it came from.

---
title: "CO-STAR Prompt Framework (CO-STAR): Template & Examples"
description: "Learn the CO-STAR framework — Context, Objective, Style, Tone, Audience, Response."
source: https://www.promptedit.app/prompt-framework/costar
---

# CO-STAR Framework
...

Endpoints

RequestHeaderReturnsNotes
GET /{path}Accept: text/markdowntext/markdownAny page, as Markdown. Returns HTML without the header.
GET /{path}.md—text/markdownSame document, for clients that cannot set request headers.
GET /llms.txt—text/plainCurated index of the site.
GET /llms-full.txt—text/plainFull text of every reference page in one document (~800 KB).
GET /sitemap.xml—application/xmlAll page URLs.
POST /mcp—application/jsonModel Context Protocol server (Streamable HTTP, JSON-RPC 2.0).
GET /.well-known/mcp/server-card.json—application/jsonMCP server card: endpoint, capabilities and tools.
GET /.well-known/agent-skills/index.json—application/jsonAgent Skills discovery index (16 skills).
GET /openapi.json—application/jsonOpenAPI 3.1 description of everything on this page.
GET /.well-known/api-catalog—application/linkset+jsonRFC 9727 catalog pointing at the spec and these docs.

Bulk retrieval

If you would rather not crawl, /llms-full.txt contains the full text of every framework guide, comparison and prompt flow in a single document, each section preceded by its source URL. /llms.txt is the short index.

MCP server

The framework reference is available over the Model Context Protocol atPOST /mcp, using the Streamable HTTP transport. It is stateless and needs no authentication. Three tools are exposed — list_frameworks,get_framework and recommend_framework — alongside every guide as a resource and a prompt per featured framework.

curl -X POST https://www.promptedit.app/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

The server card is at /.well-known/mcp/server-card.json. Note that recommend_framework matches keywords against the catalogue; it is a lookup, not a judgement, so treat its results as candidates to read.

Agent skills

/.well-known/agent-skills/index.json lists 16 downloadable skills: a router that teaches an agent to choose among all 55 frameworks, plus the 15 featured frameworks individually. Each entry carries a SHA-256 digest of its SKILL.md.

Machine-readable description

/openapi.json is an OpenAPI 3.1 description of the endpoints above. /.well-known/api-catalog is an RFC 9727 catalog pointing at both the spec and this page.

Usage terms

This content is published with Content Signals permitting search indexing, AI input and AI training — see /robots.txt. Attribution to the source URL is appreciated. There is no rate limit; the site is static and served from a CDN, but please fetch /llms-full.txt once rather than crawling 78 pages repeatedly.