Hatchdoor

What the Hatchdoor MCP server does

The Hatchdoor MCP server serves folders of Markdown notes through two interfaces: a web application for human users and an MCP endpoint for compatible AI clients. Obsidian is not required, and the vault does not need to be opened in the Obsidian desktop application. Any folder containing Markdown notes can be used as a vault.

Agents can read notes, search by keyword or meaning, create and edit files, move notes, and maintain links between notes. They can also perform checked tag changes across a vault and transfer attachments through temporary upload and download links. The browser interface supports browsing, rendering, editing, attachments, backlinks, graph views, metadata, and saved queries written with Obsidian Bases syntax.

A single installation can serve multiple vaults. Each vault can have its own source location, Git synchronization settings, and agent scope. This makes the Hatchdoor MCP server suitable for a homelab or other self-hosted setup where several independent note collections must remain separated.

How it works

Markdown files remain the authoritative data. Hatchdoor creates a disposable SQLite read model for navigation, metadata, links, backlinks, graph information, keyword search, and semantic search. Removing that cache does not remove the notes; the index can be rebuilt from the vault.

Writes from the web UI and MCP clients use the same vault operations. The project describes these operations as atomic and uses optimistic concurrency checks to reduce the risk of overwriting a newer change. Link handling covers Obsidian wikilinks and Markdown links to note files, including updates when notes are renamed or moved.

Agent responses are designed to limit context usage. Search results can be compact, long notes can be represented by outlines or selected sections, and exact-string searches can report every occurrence. Optional automatic Git commits and pushes can preserve Hatchdoor writes, while synchronization conflicts are handled through a recovery branch on the Git host.

Setup and configuration

Hatchdoor is intended to run as a self-hosted service. The README documents Docker Compose as an approachable installation path and also identifies Docker and Podman deployment options. Its published container image is rootless and distroless, running as the nonroot user rather than providing an interactive shell.

A fresh installation includes a setup checklist that helps connect a vault and reach agent search. The interface includes a folder picker, so configuration does not require users to enter container paths manually. Documentation is available inside the application, through MCP documentation tools, and as Markdown under /docs/ on an instance.

The project supports optional Git-backed vault synchronization and per-vault settings. Exact environment variable names and authentication configuration are not specified in the supplied material, so they should be taken from the project documentation before deployment.

Tools and capabilities

The Hatchdoor MCP server supports these agent-facing tasks:

• Read notes and retrieve focused sections from long documents.
• Search notes with keyword or semantic queries.
• Find exact text occurrences and their counts.
• Create, edit, move, and link notes.
• Rename or remove a tag across a vault as one checked operation.
• Upload and download attachments using short-lived links.
• Read and search the application documentation through read_docs and search_docs.

The server is a fit for private Markdown collections that need both browser access and agent access. It is not presented as a hosted synchronization service, a multi-user collaboration platform, or a replacement for Obsidian. Its model is a self-hosted companion for vaults that the operator controls.