The human interface for AI

Give your agents a workspace humans understand.

Astro brings instructions, context, tools, and outputs into one self-hosted interface. You work in Markdown, messaging, tables, and dashboards. Your agents work through MCP. Both work on the same things.

Install Astro Explore the docs

curl -fsSL https://runastro.sh/install.sh | bash

Open source · Runs in Docker · Your data stays yours

Product launch · Astro

Human-readable instruction

MARKDOWN

Prepare the launch brief

Review our research, compare the alternatives, and update the launch dashboard.

MCP search documents

MCP update dashboard

Agent-readable tools

01read_markdownInstructions loadeddone

02search12 sources founddone

03upsert_dashboardWriting inspectable outputlive

Output visible to everyone

▦ Dashboard updated

▤ Comparison table created

One shared workspace across

MarkdownMessagingMCPRESTLocal search

Stop rebuilding the glue

Your AI workflow already has an interface.
It is just scattered everywhere.

Prompts live in Markdown. Tasks happen in messaging. Context sits in folders and vector databases. Outputs disappear into message threads. Astro turns those pieces into a coherent system without asking you to abandon the tools you already use.

WITHOUT ASTRO

A fragile pile of integrations

01Copy prompts between files and chat
02Build another one-off retrieval layer
03Wire cron, messaging, and scripts together
04Dig through agent logs for the result

WITH ASTRO

One legible operating layer

01Author durable instructions in Markdown
02Give agents context and actions through MCP
03Coordinate recurring work in messaging
04Inspect results in shared, structured views

One corpus, two interfaces

Designed for people.
Native to agents.

Astro gives humans a visual workspace and agents a broad MCP surface over the exact same objects. There is no shadow memory to reconcile and no special agent-only version of your project.

See everything agents can use →

M↓MarkdownsInstructions and durable context DDocumentsSearchable source material ▦TablesTyped, structured outputs ◇DiagramsAgent-editable Excalidraw >_ScriptsRunnable Python workflows ▥DashboardsInspectable shared status

The collaboration loop

From human intent to inspectable output.

Astro makes the handoff between people and agents explicit, repeatable, and easy to understand.

✎

Write the intent

Create reusable instructions in Markdown. Insert concrete MCP tool calls and object IDs without memorizing syntax.

↗

Send the work

Run now, schedule once, or use cron. Agent Tasks deliver the instruction into the messaging channel where collaboration happens.

⌁

Let agents act

Any MCP-compatible client—Cursor, Claude Desktop, OpenAI Codex, Hermes, or your own agent—can search, read, write, organize, and run workflows against your Astro instance.

✓

See what changed

Results land in dashboards, tables, diagrams, markdown, or captured task logs—not buried in a disposable conversation.

How it fits together

Messaging, agents, and Astro in one loop.

Astro sits in the middle: humans author and inspect work there, messaging platforms coordinate delivery, and MCP clients act on the same shared corpus.

Humans author in Astro. Agent Tasks reach messaging platforms such as Slack and Discord. MCP clients like Cursor, Claude Desktop, OpenAI Codex, and Hermes read and write the same Astro objects.

Built for real experiments

The pieces your agent stack keeps needing.

Useful primitives for building serious workflows, packaged in one self-hosted application.

⌘CORE

60+ MCP tools

Read, write, search, organize, schedule, and execute across Astro from Cursor, Claude Desktop, OpenAI Codex, Hermes, or your own agent.

searchread_markdownwrite_table_rowrun_scriptupsert_dashboard

◎

Category canvases

Pan, zoom, and organize markdown, documents, diagrams, tables, scripts, and links in a spatial workspace.

Messaging Agent Tasks

Turn Markdown into scheduled instructions for supported messaging platforms, with channel delivery, mentions, run history, and configurable templates.

Python automation

Save, run, and schedule Python with captured stdout, stderr, status, and execution history.

◌

Portable Universes

Separate experiments by project, keep their context clean, and export or restore your work when needed.

M↓

MCP-aware Markdown

Write durable agent instructions and insert concrete MCP tools, Universe context, and object IDs directly from the editor.

⌕

Local search and memory

Search across markdown, documents, scripts, diagrams, tables, and links with fast text retrieval and optional local embeddings.

▥

Agent-editable dashboards

Give every Universe a shared status surface where agents can publish widgets and link full Markdown reports.

▦

Typed tables

Keep structured data legible with typed columns, row-level MCP tools, inline editing, search, and CSV import or export.

◇

Excalidraw diagrams

Create visual artifacts humans can edit and agents can read, write, and update as native Excalidraw data.

⌂

Self-hosted by default

Run in Docker with persistent volumes, local embeddings, optional Tailscale access, and data under your control.

Connect your agent

One endpoint.
A shared world of context.

Point any MCP-compatible client at Astro. Your agent gets tools for the same workspace you see and edit.

CursorClaude DesktopOpenAI CodexHermesCustom agents

Read the MCP docs

mcp.json

{
  "mcpServers": {
    "astro": {
      "url": "https://your-host/mcp/"
    }
  }
}

Connected · tools available

See it in action

Go from zero to a shared agent workspace.

Human intent. Agent execution.

Build the interface between them.

Bring your instructions, context, automations, and outputs together in a workspace both sides can use.

Install Astro View on GitHub Get the extension