Choose an agent integration

For shell automation, use the agmd CLI. For an MCP client or a custom application, use the matching integration in the following table:

Integration	Use case	Setup
Share URL	One agent working on one document	Give the agent a scoped link
agmd CLI	Shell workflows, scripts, coding agents	Install one executable
MCP	Assistants that discover and call tools	Configure the remote MCP endpoint
HTTP API	Applications and custom orchestration	Use documented routes and versions

All four operate on the same canonical Markdown, roles, comments, suggestions, revisions, and event history.

Start with a share URL

For a bounded task, create a distinct suggest or edit link and pass the complete URL to the agent.

Ask the agent to:

1. Read before writing.
2. Preserve the document structure.
3. Attribute its work.
4. Use suggestions when review is required.
5. Handle a version conflict instead of forcing an overwrite.

Use the CLI for shell work

To install the CLI and read a document, run these commands:

curl -fsSL https://agentsmarkdown.com/install.sh | sh
DOC='https://agentsmarkdown.com/d/DOC_ID?key=SHARE_KEY'
agmd docs cat "$DOC"
agmd comments add "$DOC" "Ready for review." --line 12
agmd watch "$DOC" --json

Replace DOC_ID and SHARE_KEY with the values from your share link. For more information, see Use the CLI.

Use the Model Context Protocol

The Model Context Protocol (MCP) endpoint provides document and Node media tools:

https://agentsmarkdown.com/mcp

Continue with MCP.

Use HTTP for an application

The HTTP API provides the complete document contract and machine-readable OpenAPI:

• OpenAPI JSON
• Agent skill
• Quickstart
• Agent index

Continue with HTTP API.

Preserve least privilege

Give each participant a separate credential and the narrowest useful role. A reviewer usually needs suggest, not edit. A monitor usually needs only view.

Never place an owner share key or account session token in a public document, command argument that will be logged, or source control.