NoteDiscovery Adds HTTP MCP Server for AI Clients

The maintainers of NoteDiscovery, an open-source notes project, have released version 0.32.0, which includes work on an HTTP-based Model Context Protocol (MCP) server. The change closes a feature request on the project's GitHub repository that users had pushed for through comments from August to October. For people who self-host their notes and want AI assistants to work with them, the request addressed a real obstacle.

What the user wanted

The request came from a user who keeps a personal knowledge base in Obsidian and wanted a lightweight, headless service on a network-attached storage (NAS) device. The goal was to let a locally run language model query and edit those notes through MCP.

According to the issue, the user found the existing MCP setup difficult to configure within open-webui's mcpo proxy. The user proposed a simpler design: a single environment variable in the Docker Compose file would switch on an HTTP server listening on a separate port. Ideally, the web interface and the MCP service would share one container, although the user said two containers would also be acceptable.

The user also described a second problem. The MCP setup at the time launched through a Docker command, which required the Docker command-line client and access to the Docker socket. The user considered that arrangement unsafe and impractical for several reasons.

Why the setup mattered

MCP is an open standard that lets AI applications connect to external tools and data sources. Many early implementations ran as local processes that exchanged messages through standard input and output. That approach suits a single desktop, but it is awkward for a server that several clients need to share.

The user's end goal was an AI agent able to explore the link graph of the notes, including links, tags, keywords and folders, and to create, read, update and delete items. The issue also noted a usability gap. In Obsidian, adding a tag is quick and applies across the vault, while in NoteDiscovery the tag had to be stored in each note's metadata.

Follow-up requests

Other users joined the discussion in August and October. One asked whether the service could work directly with web-based AI products, with authentication. Another user said they wanted to register the server with an MCP gateway, a tool that combines several MCP servers behind one endpoint so that clients such as open-webui or opencode can use them without separate configuration.

Those commenters described the desired transport as streamable HTTP with token authentication, reachable at a fixed path on a container and port. Gateway setups like this are common among people who run several AI tools at home, because they avoid repeating configuration in every client.

From pull request to release

On October 10, a participant said they had created pull request #314 and that its tests passed. The issue was closed as completed through that pull request on October 13. The project owner, known as gamosoft on GitHub, then announced that release v0.32.0 was available and thanked users for their feedback and support.

The issue thread itself does not list every change in the release, so readers who need details on configuration options, authentication or supported endpoints should check the release notes and the documentation in the repository.

Why it matters for self-hosted users

For users of self-hosted software, the difference between a local helper process and a network service is practical. A network endpoint can run on a home server, be reached from several devices, and sit behind a gateway. It also avoids giving a container control of the Docker daemon, which is the concern the original author raised.

Self-hosters can run the notes service and its AI interface on the same machine without installing extra local tooling.
Gateways can offer one MCP endpoint to many clients, reducing repeated setup.
Removing the need to expose the Docker socket lowers the security risk of the arrangement.

The episode also shows how user feedback in a public issue tracker can shape a project. The original report was framed as a note for others who might run into the same problems, and it prompted several users to add their own requirements.

Outlook

The release suggests that NoteDiscovery is moving toward a model where its notes can be used by AI agents over standard network protocols. Users who want to test the feature should read the v0.32.0 release notes first, since the issue discussion alone does not confirm every detail of what shipped. Interest in combining personal notes with local language models is likely to keep growing, and this change puts NoteDiscovery closer to that use case.