Documentation
Nibvine has one data model and two ways into it: a REST API for anything that speaks HTTP, and an MCP server for AI clients. Both are generated from the same project tree and enforce the same scopes.
- Quickstart From an empty account to an agent reading your project, in five steps with copy-pasteable commands. Start here
- REST API Every route the server exposes, with methods, authentication and scoping rules. 195 endpoints
- MCP server Every tool an AI client can call, its arguments, the scopes it needs, and how safe writes work. 55 tools
How the pieces fit
-
A project is a tree of notes
Each note has a parent, a depth and an order. Tasks, ideas and comments attach to the project or to a specific note. Everything you read through either interface comes out of that structure.
-
A token decides what is visible
Tokens carry scopes, and may be restricted to a single project. A scope you did not grant is a call that gets refused, on both interfaces.
-
A context pack is the model-facing view
Rather than fetching a whole tree, an agent can ask for a context pack: the project assembled into one payload, summarised where a branch is too large for the window.
-
A write happens twice
External writes are previewed first and committed second. The preview describes exactly what would change and expires if nothing commits it.