md-manager
Docs for humans & AI agents

Documentation your agents can actually use.

md-manager keeps your markdown in Postgres — versioned, searchable, and permissioned — and lets people and AI agents read and write it under the exact same rules. One source of truth, three ways in.

Start with GoogleExplore the docs

web app · CLI · MCP server · OAuth connectors

handbook / runbooks / deploy.md

Deploy Runbook

Ship a release to production. Owned by @platform, kept current by both humans and the on-call agent.

  1. Tag the release and wait for green CI.
  2. Promote with mdm ship --canary.
  3. Watch error rate; roll back on regression.
agent › get_doc("deploy") → 3 steps, v7, updated 2m ago
For people

A clean place to write

A focused editor with live preview, version history, full-text search, teammates, and roles. Sign in with Google and you're in.

For agents

A clean place to act

The same documents over a CLI and an MCP server — create, update, append, search, tag — with scoped keys and the identical permission model.

Why md-manager

Everything a shared knowledge base needs — for both kinds of readers.

Agent-native

Built for agents, not bolted on

An MCP server and a CLI expose the same 20 operations, so Claude, Gemini, and GPT read and write your docs directly — with the same permissions as a person.

Safe writes

Conflict-aware editing

Every save is version-checked. A stale write returns the current and base content so humans or agents can do a clean 3-way merge instead of clobbering work.

Findable

Search that actually finds it

Postgres full-text out of the box, plus semantic and hybrid search with pgvector — aggregated to documents, not raw chunks.

Multi-tenant

Teams, enforced in the database

Organizations, teams, projects, and a real role lattice — isolated by Postgres row-level security, not just app-layer checks.

Versioned

Full history, instant restore

Every version is a snapshot. Roll back any document to any point. Agent autosave churn is coalesced so history stays readable.

Yours

Lives in Postgres

Your documents are rows in a database you control — never files scattered across laptops and chat threads. Export, back up, and own them.

In practice

Two lines for a person. Two tools for an agent.

The CLI is great for shell-capable agents (cheap, scriptable). The MCP server is for hosts that speak MCP. Both go through the same API and rules.

CLIread & write from any shell
terminal
# install once, then point it at this instance
mdm auth login --api-url https://mdm-api.nimy.work --api-key mk_live_…

# create a doc, then read it as raw markdown — pipe straight into an agent
mdm doc create --project handbook --path runbooks/deploy \
  --title "Deploy" -m "# Deploy\n1. Tag the release\n2. Ship"
mdm doc get 019f… | claude -p "summarize the deploy steps"

# search across everything (keyword, semantic, or hybrid)
mdm search "rollback procedure" --mode hybrid
MCPpoint an MCP host at your docs
claude_desktop_config.json
// point any MCP client at the server with a key — no install
{
  "mcpServers": {
    "md-manager": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mdm-api.nimy.work/mcp",
               "--header", "Authorization: Bearer mk_live_…"]
    }
  }
}
CLI reference →MCP setup →Web connectors →

Bring your docs into one place your team and your agents both trust.

Free to start. Sign in with Google, create your organization, and invite the rest — humans and agents alike.

Start with GoogleRead the docs