$mdflowmarkdown-ai
$mdflow --agent

> markdown your AI agents can read & write

Every LLM already speaks Markdown. MDflow goes further — it turns a whole workspace into something Claude Code, Codex, Cursor, and Gemini operate directly: read context, write documents, organize folders, and publish, all through a remote MCP server and a full HTTP API.

./start-freeconnect an agent →
agent-session.log
$claude
# mdflow MCP connected · 16 tools
 
> "summarize our onboarding docs and draft a checklist"
 
mdflow_get_context("onboarding")
  ↳ 4 docs · folder: Onboarding
mdflow_create_document("onboarding-checklist.md")
  ↳ created in Onboarding/
mdflow_update_document_sharing(public: true)
  ↳ https://mdflow.cz/share/k3f9…
 
✓ done — a clean shared page + its .md twin
$why markdown

every model already speaks markdown

Plain text with just enough structure. Headings, lists, and code fences carry meaning a model can parse — so instead of guessing how your content is organized, the structure is explicit. That's why every major model reads and writes it natively.

#
Headings
become document structure
-
Lists
become enumerable steps
```
Code fences
become verbatim context
|
Tables & links
become data and relations
$ls /.well-known

agents can discover MDflow on their own

MDflow publishes machine-readable files so a capable agent can find the docs, the control surface, and the rules — without a human wiring it up.

/llms.txt

Discovery index — points agents to the docs, API, MCP server, and pricing.

/docs.md

Self-contained control manual — auth, every MCP tool, and the REST endpoints on one page.

/pricing.md

Machine-readable plan limits, so an agent can tell what a workspace allows.

/.well-known/agent-card.json

A2A discovery beacon — interfaces, security scheme, and skills.

<share-url>.md

Append .md to any share URL for YAML frontmatter plus the body in one request.

$use-cases

what teams build on agent-controlled markdown

$mdflow_get_context

A knowledge base agents can read

Folder descriptions give scoped context, and get_context retrieves the right documents on a topic — retrieval without a separate vector store.

$mdflow_create_document

Agents that write for you

Meeting notes, specs, and changelogs created straight into the right folder — not pasted back for you to file by hand.

$cat prompts/*.md

Prompt & context libraries

Keep reusable prompts, specs, and style guides as Markdown your agent pulls on demand, every run.

$mdflow_update_document_sharing

Publish agent output

An agent writes a document, flips on a public link, and hands a human a clean page — or its .md twin for the next agent.

$how mdflow_get_context works
  1. 1Rank folder descriptions — the primary signal.
  2. 2Then folder names and document titles.
  3. 3Fetch the best-matching bodies — up to 10 docs, 50k chars each.
$cat pricing.txt

agent access on Pro

Writing and the editor are free. Agent control — the HTTP API, the remote MCP server, and Personal Access Tokens — is part of Pro: €4.99/month with a 7-day free trial.

Freefree forever
  • [x]5 markdown files
  • [x]5 image uploads
  • [x]Public sharing links
  • [x]Commenting
Pro€4.99/ month
  • [x]Unlimited markdown files
  • [x]10,000 image uploads
  • [x]Full HTTP API access
  • [x]Remote MCP server
// no card to start · fair-use applies
./start-freefull pricing
$cat faq.md

markdown for AI agents — FAQ

Why is Markdown the best format for AI agents?
Every major model — GPT, Claude, Gemini, Llama — reads and writes Markdown natively. Its headings, lists, and code fences carry explicit structure, so the model parses meaning instead of guessing. And it's plain text, so it stays portable across any tool or pipeline.
Which AI agents and tools work with MDflow?
Any MCP-capable client — including Claude Code, Claude Desktop, Cursor, Codex, and the OpenAI Responses API — connects to the hosted remote MCP server. Anything that can send an HTTP request can use the REST API.
Can an AI agent create and edit my documents, or only read them?
Both. With a Personal Access Token an agent has full read and write — retrieve context, create, update, move, and delete documents and folders, and manage sharing. The same surface you have.
What is the MDflow MCP server?
A Model Context Protocol server at https://mdflow.cz/api/mcp exposing 16 tools for your workspace. It's hosted and remote (Streamable HTTP), so there's nothing to install — a local stdio server is available too.
Do I need to run anything locally?
No. The remote MCP server is hosted — connect with one line and a token. A local stdio server is optional, for clients that prefer to launch a process.
How does an agent know what's in my workspace?
Folder descriptions are the primary context signal. mdflow_get_context ranks them first, then titles, then returns the best-matching document bodies — topic retrieval without a separate vector store.
Is agent access free?
The editor and writing are free — 5 documents and 5 images. Agent control (the API, the MCP server, and Personal Access Tokens) is part of Pro: €4.99/month with a 7-day free trial.
Is it safe to let an agent write to my workspace?
Tokens are scoped to your workspace, can be revoked instantly, and are rate-limited to 30 requests per minute. Ownership is enforced on the server, and a token sent as a tool argument is refused — it must live in client config, never in a prompt.
$mdflow init

Point your agent at a workspace built for it.

Sign in with Google, create a folder, and write Markdown that people and AI agents can read. The free plan needs no card; agent access is on Pro.

./start-free →read the agent guide