kb-mcp
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@kb-mcpsearch for lessons about MCP integration"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
kb-mcp
An agent-native knowledge base.
pip install kb-mcp โ give any LLM agent a structured, queryable, local-first second brain.
The problem
Knowledge bases for humans (Notion, Obsidian) and for search engines (Elasticsearch, vector DBs) leave a gap: LLM agents need a knowledge layer that speaks their protocol and assumes the reader is a model, not a person.
kb-mcp fills it.
Obsidian / Notion | Vector DBs (Chroma / LanceDB) |
| |
Reader-optimised for | Humans | Embeddings | LLM agents |
Protocol | Web UI | SDK | MCP (stdio) |
Schema | Free-form | Free-form | Typed (project / decision / lesson / ...) |
Default storage | Cloud / proprietary | Local files | SQLite + FTS5 |
Setup | Sign up |
|
|
Related MCP server: vault-memory
Features
๐ง Agent-native. Every document is reachable from any MCP client (
Claude Desktop,Cursor,OpenCode,Codex, โฆ) viakb_search/kb_get/kb_add/kb_link.๐ Schema-first. Six built-in document types (
project,decision,lesson,glossary,person,faq) โ extensible via Python subclassing.๐ Full-text search. SQLite FTS5 with BM25 ranking. Snippet-aware results returned to the agent.
๐ Typed links. Documents reference other documents; backlinks are automatic.
๐ Markdown friendly. Round-trip import/export with frontmatter. Humans can edit, agents can read.
๐ชถ Zero deps by default. SQLite ships with Python.
pip install kb-mcpand you're done.๐ Local-first. Your data lives in
~/.local/share/kb-mcp/. No cloud, no telemetry, no phone-home.
Quickstart
pip install kb-mcp
kb init
kb add --type project --title "kb-mcp" --tags kb,mcp,open-source --body "Agent-native knowledge base."
kb search "mcp server"
# Expose to any MCP client
kb serveThat's it. Five commands, zero config files.
๐ Full walkthrough: docs/quickstart.md
Document types
Type | Purpose | Example |
| Repo / initiative background |
|
| Architecture Decision Record (ADR) | "Use SQLite FTS5 over Elasticsearch" |
| Post-mortem / lessons learned | "Don't |
| Term definitions |
|
| People the agent should recognise | "Zhang Bei, owner, uses Hermes" |
| Frequently asked questions | "Why SQLite?" |
Subclass kb_mcp.schema.Document to add your own.
MCP integration
Add to ~/.config/claude_desktop_config.json (or any MCP client):
{
"mcpServers": {
"kb": {
"command": "kb",
"args": ["serve"]
}
}
}The agent then sees four tools:
kb_search(query, type?, tags?, limit?)โ BM25-ranked results with snippetskb_get(id)โ full document by id (or slug)kb_add(type, title, body, tags?, source?)โ create documentkb_link(from_id, to_id, rel?)โ typed edge between documents
Development
git clone https://github.com/your-org/kb-mcp
cd kb-mcp
pip install -e ".[dev]"
pytest # unit + E2E (real SQLite temp file, no mocks)
ruff check .
mypy src/๐ Spec: docs/requirements.md ยท Architecture: docs/architecture.md ยท CLI reference: docs/cli-reference.md
Roadmap
Version | Scope | Status |
v0.1.0 | CLI + MCP server + SQLite/FTS5 + 6 doc types + Markdown I/O | ๐ง in progress |
v0.2.0 | Vector search (sqlite-vss) as opt-in, hybrid BM25 + embedding ranking | planned |
v0.3.0 | Multi-vault (per-project isolated KBs) + shared-vault mode | planned |
v0.4.0 | Web UI (read-only) + collaborative editing hints | exploring |
v1.0.0 | Postgres backend, multi-user auth, hosted mode | exploring |
See docs/requirements.md ยง 4 for v0.1 scope decisions and out-of-scope list.
Status
alpha. API and storage format may change before v0.2.0. Pin minor versions
(kb-mcp>=0.1,<0.2) in production.
Contributing
Issues and PRs welcome. See CONTRIBUTING.md (TODO before v0.1.0 release).
By participating, you agree to abide by the Code of Conduct (TODO before v0.1.0 release).
License
MIT โ do what you want, just keep the copyright notice.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/HelloTomBruce/kb-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server