junnnnnw00/obsidian-everywhere

obsidian-everywhere is a TypeScript server that lets AI tools like Codex, ChatGPT, and Claude read and edit an Obsidian notes vault through the Model Context Protocol, or MCP. Instead of treating the vault as a plain folder of markdown files, the server represents notes and the links between them as a graph, so an AI client can ask for things like backlinks, neighboring notes within a certain number of hops, or a bundle of related context around a topic, rather than only reading raw file text. Under the hood it parses the markdown files, including wikilinks, embedded content, frontmatter, nested tags, headings, and block references, and keeps two things in sync as the vault changes: a SQLite index with full text search, and an in-memory graph structure that supports multi-hop traversal, shortest path lookups, and PageRank style ranking of important notes. The README states this index updates incrementally rather than being rebuilt from scratch each time. On top of this, the project exposes 31 separate tools an AI client can call, split between reading tools such as searching notes, listing backlinks, finding orphaned or unresolved notes, and writing tools such as creating, moving, renaming, or deleting notes, updating frontmatter, and running bulk find and replace operations with a rollback option. The server can connect to clients in three ways: through standard input and output for local tools like the Codex CLI, over HTTP with a fixed bearer token for private remote use, or over HTTP with OAuth 2.1 for public connectors. Write tools that can change the vault are turned on by default for local and bearer token connections, but turned off by default for the public OAuth connection unless specifically enabled. The README includes a demo command that builds a temporary sample vault to show off the graph features without touching a real one, plus setup commands that print ready to paste configuration for Codex, ChatGPT Desktop, and Claude, and a doctor command that checks the environment for problems. The project is released under the MIT license.