Semiont MCP
A knowledge base from your documents where people and AI agents annotate together, with an MCP server and skills for Claude Code
- 93 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The MCP server reads and writes annotations as the signed-in user
- The stack runs containers and keeps secrets, by default in plain files
- In cloud mode document text goes to the model provider
Install
- Works in
Manual install
brew install the-ai-alliance/semiont/semiontHomebrew installs the launcher on macOS and Linux. For Windows the releases carry a zip with `semiont.exe`. Docker, Podman or Apple Container is required.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The MCP server reads and writes annotations as the signed-in user
- The stack runs containers and keeps secrets, by default in plain files
- In cloud mode document text goes to the model provider
This is third-party code. Review the repository files before installing.
What it does
Semiont turns documents into a knowledge base in which every claim is pinned to a passage of its source. A reader highlights text, adds tags and links a claim to the passage that supports it, while AI agents add references and entity types in the same annotations. Each record keeps who made it, a person or an agent. Everything is run by the `semiont` program, a static binary with no npm or Node.js. `semiont init` creates a knowledge base in the current folder, and `semiont start` brings up the stack in containers (Apple Container, Docker or Podman) and opens the browser interface at `localhost:3000`. AI jobs need an Anthropic key or a local model through Ollama. Semiont reaches agents in two ways. `semiont mcp` starts an MCP server over stdio: after `semiont login`, Claude Code or another MCP client reads the knowledge base and writes to it as the signed-in user. The tools are named after the protocol's eight verbs: `browse_`, `match_`, `gather_`, `mark_`, `yield_` and others. The second part is twelve skills in `docs/builder/skills/` that teach an assistant to write programs on the SDK: ingestion, highlights, comments, assessments, tags, relations and aggregates.
Who it is for. Analysts and research teams that accumulate documents and want answers traceable to a passage, and developers who connect Claude Code to such a base.
Good fit when
- A corpus of contracts or papers has to be annotated so each claim leads back to its source
- An agent finds links between entities and a person accepts or rejects each one
- Claude Code should search and extend notes in the base through MCP, under your account
- The knowledge base has to live on your server or laptop without documents leaving for a cloud
- You need aggregates: a report or a timeline assembled from annotations on one subject
Not a fit when
- You want personal notes with no containers or sign-in: something simpler such as Obsidian fits better
- You need MCP as a remote endpoint: the server runs on your machine over stdio
- There is no container runtime and none can be installed: the Semiont stack runs in Docker, Podman or Apple Container
- You need a stable SDK: it is in alpha now and APIs may change between 0.x releases
Example request
Upload the papers folder into the Semiont base and mark every mention of quality metrics as a reference to the Metric entityMCP
- Transport
- stdio
- Authentication
- not required
Limitations
- A container runtime is required: Apple Container, Docker or Podman.
- AI suggestions work with an Anthropic key or a local Ollama model. Secrets are kept in plain files by default, for development; a 1Password store is a setting away.
- The SDK and API are in alpha, and breaking changes between 0.x releases are expected.
- The MCP server serves only your machine over stdio and acts as the signed-in user; an agent has no separate account.
How to disable. `claude mcp remove semiont` disconnects the server in Claude Code. `semiont stop` shuts the stack down and `semiont logout` ends the launcher session. Skills are removed by deleting their folders from `.claude/skills/`.
FAQ
How do I connect Semiont to Claude Code?
Run `semiont login`, then `claude mcp add semiont -- semiont mcp`. For another MCP client, add a server entry with the command `semiont` and the argument `mcp`.
What does the Semiont MCP server do?
It reads the base (`browse_`, `match_`, `gather_`), writes annotations and resources (`mark_`, `bind_link`, `yield_`, `frame_`), tracks jobs (`job_status`, `job_cancel`) and serves the base's resources as MCP resources.
Does Semiont work without a cloud?
Yes. The stack runs locally, and instead of Anthropic you can pick Ollama with `semiont init --inference ollama`, in which case content stays on the machine.
How do the Semiont skills differ from the MCP server?
The skills help an assistant write a program on the SDK, while `semiont mcp` lets an AI application work in the base itself.
Which languages can I use with the SDK?
The docs name TypeScript, Rust, Python, Go, Kotlin with Java, C#, Swift and Haskell.
Discussion
No questions or reviews yet.
Related
MCP serversSemantica plugin: 17 skills and 3 agents for knowledge graphs, reasoning and data provenance in Claude Code, Cursor and Codex.
A local MCP server that turns a folder of documents into a knowledge base with hybrid search for Claude Code and Cursor
A knowledge graph with an MCP server: the agent searches concepts by meaning and sees how well each is supported.
A CLI and MCP server that extracts structured knowledge from documents: graphs, hypergraphs, and fact search with sources
A self-hosted RAG app with a web interface and a built-in MCP server: documents, Qdrant and answers with sources.
InfraNodus MCP server turns any text into a knowledge graph and shows topic clusters, the gaps between them and research questions
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Semiont (this card) | no | not required | 93 | |
| Basic Memory | no | not required | 4,566 | agent memory in local Markdown files, without claims pinned to passages or shared annotation |
| claude-obsidian | no | 16,397 | a knowledge base in an Obsidian vault, while Semiont keeps annotations with an author per record in its own stack |
README in short
The README briefly explains the idea of a knowledge base with claims tied to sources and gives a Homebrew install command. Below are four documentation tracks: analyst, operator, builder and contributor. Details live in the Quick Start and in the guide to connecting AI applications.