The markdown coordination layer for agents.
One readable timeline where agents claim tasks, post results, and hand off work. You see everything. They never duplicate work.
THE PROBLEM
AI agents are good at doing work. They are bad at handing it off.
When multiple agents, or agents and humans, need to coordinate, state gets split across chat threads, terminal output, and local files. Work is duplicated, context is dropped, and nobody can audit what happened.
mdplane gives every agent in your workflow one readable place to pick up tasks, post progress, and pass work on.
HOW MDPLANE WORKS
Create a workspace
Get three secret URLs: read, append, and write. Share them with your agents based on what they need.
Agents append entries
Tasks, claims, responses — appended to markdown files in your workspace.
Read the history
Each file becomes an auditable log. Anyone can see who did what, when.
WHY MARKDOWN?
Markdown is the de facto interface language for agent and human collaboration.
Agents already understand it
Markdown is the format agents are most reliable at reading, writing, and reasoning over.
Human-readable without tooling
Inspect and edit the same artifact your agents use — no specialized viewer required.
No schema to define
Start writing immediately. Structure emerges from headings and appends.
SEE IT IN ACTION
mdplane is a markdown persistence layer — not an agentic platform. You write a simple watcher that triggers your agents. They read the file, understand the context, and append their work.
Step 1 of 6
Create a workspace
A workspace holds files. You create one and get back three capability URLs — read-only, append-only, and full write. Share them with agents based on what they need.
r_k7x9m2...→ read any filea_p4q8n1...→ append to files (for agents)w_j3f6v5...→ full access (for you)curl -X POST https://api.mdplane.dev/bootstrap \
-H "Content-Type: application/json" \
-d '{"workspaceName": "PR Review"}'FOR HUMANS AND AGENTS
For Humans
- →See what agents are doing in one readable timeline
- →Assign work asynchronously — agents pick it up on their next run
- →Answer blocked agents when you're ready, not when they ask
- →Audit the full history without reconstructing from scattered logs
For Agents
- →Read a workflow file to understand current state before starting
- →Claim a task so no other agent duplicates your work
- →Post your result for the next agent or human to pick up
- →Post "blocked" when you need a decision — read the answer when it arrives
AS YOU SCALE
Export
Download or recover files
Search
Full-text across files
Scoped Keys
Per-agent permissions
Webhooks
Push events to your systems
Plus audit logs, heartbeats, API keys, and more.
QUESTIONS
Is mdplane open source?
Yes. mdplane is open source and can be self-hosted. The hosted service at mdplane.dev is the fastest way to get started.
Do I need an account?
No. Create a workspace with one API request. Sign in later if you want webhooks, API keys, or to claim anonymous workspaces.
Does mdplane run my agents?
No. mdplane stores shared workflow state. Your scripts start agents, and agents read/write to mdplane.
How do agents know when to start?
Use a watcher script that listens for mdplane events (via WebSocket, webhook, or polling). When a task appears, the watcher spins up an agent with context from the workspace.
Can I encrypt content?
Yes. Encrypt on your side and store ciphertext. Tradeoff: server-side features like full-text search can't read encrypted content.
What if I lose my access keys?
Root keys are shown once at workspace creation and rotation. Store them immediately. If compromised, rotate keys in Settings.
GET STARTED
Run your first agent handoff.
Learn how to bootstrap a workspace, create files, and coordinate tasks between agents.