ContextFS
Universal AI memory layer that provides cross-client, cross-repo context management with semantic search, automatic code indexing, and session management. Enables persistent developer memory across projects with typed memories, graph-based relationships, and RAG-powered retrieval.
README
ContextFS
Universal AI Memory Layer - Cross-client, cross-repo context management with RAG.
Works with Claude Code, Claude Desktop, Gemini CLI, Codex CLI, and any MCP client.
Documentation | Developer Memory Workflow Guide | GitHub
Features
- Semantic Search - ChromaDB + sentence-transformers for intelligent retrieval
- Auto Code Indexing - Automatically index repositories for semantic code search
- Dual Storage - Smart routing between FTS (keywords) and RAG (semantic)
- Cross-Repo Memory - Memories track source repository automatically
- Session Management - Automatic capture and replay of conversation context
- MCP Server - Standard protocol for universal client support
- Plugins - Native integrations for Claude Code, Gemini CLI, Codex CLI
- Web UI - Browse and search memories with side-by-side FTS/RAG comparison
Quick Start
# Run with uvx (no install needed)
uvx contextfs --help
uvx contextfs-mcp # Start MCP server
# Or install with pip
pip install contextfs
# Or install with uv
uv pip install contextfs
# Or install from source
git clone https://github.com/MagnetonIO/contextfs.git
cd contextfs
pip install -e .
Upgrading
# Upgrade with pip
pip install --upgrade contextfs
# Upgrade with uv
uv pip install --upgrade contextfs
# Upgrade with uvx (automatic on next run)
uvx --upgrade contextfs --help
Usage
CLI
# Save memories
contextfs save "Use PostgreSQL for the database" --type decision --tags db,architecture
contextfs save "API uses snake_case keys" --type fact --tags api,style
# Search
contextfs search "database decisions"
contextfs search "api conventions" --type fact
# Recall specific memory
contextfs recall abc123
# List recent
contextfs list --limit 20 --type decision
# Sessions
contextfs sessions
Python API
from contextfs import ContextFS, MemoryType
ctx = ContextFS()
# Save
ctx.save(
"Use JWT for authentication",
type=MemoryType.DECISION,
tags=["auth", "security"],
)
# Search
results = ctx.search("authentication")
for r in results:
print(f"[{r.score:.2f}] {r.memory.content}")
# Get context for a task
context = ctx.get_context_for_task("implement login")
# Returns formatted strings ready for prompt injection
MCP Server
Add to your MCP client config (Claude Code, Claude Desktop):
{
"mcpServers": {
"contextfs": {
"command": "uvx",
"args": ["contextfs-mcp"]
}
}
}
Or with Python directly:
{
"mcpServers": {
"contextfs": {
"command": "python",
"args": ["-m", "contextfs.mcp_server"]
}
}
}
MCP Tools:
| Tool | Description |
|---|---|
contextfs_save |
Save memory (auto-indexes repo, logs to session) |
contextfs_search |
Semantic search with cross-repo support |
contextfs_recall |
Get specific memory by ID |
contextfs_list |
List recent memories |
contextfs_update |
Update existing memory content, type, tags, or project |
contextfs_delete |
Delete a memory by ID |
contextfs_init |
Initialize repo for auto-indexing (opt-in) |
contextfs_index |
Index current repository for code search |
contextfs_index_status |
Check or cancel background indexing progress |
contextfs_list_indexes |
List all indexed repositories with stats |
contextfs_list_repos |
List all repositories with memories |
contextfs_list_tools |
List source tools (claude-code, claude-desktop, etc.) |
contextfs_list_projects |
List all projects |
contextfs_sessions |
List sessions |
contextfs_load_session |
Load session messages |
contextfs_message |
Add message to current session |
contextfs_update_session |
Update session label or summary |
contextfs_delete_session |
Delete a session and its messages |
contextfs_import_conversation |
Import JSON conversation as episodic memory |
MCP Prompts:
| Prompt | Description |
|---|---|
contextfs-save-memory |
Guided memory save with type selection |
contextfs-init-repo |
Initialize repo for auto-indexing |
contextfs-index |
Index repository for semantic search |
contextfs-session-guide |
Instructions for session capture |
contextfs-save-session |
Save current session |
Plugins
Claude Code
# Install hooks for automatic context capture
python -c "from contextfs.plugins.claude_code import install_claude_code; install_claude_code()"
Gemini CLI / Codex CLI
from contextfs.plugins.gemini import install_gemini
from contextfs.plugins.codex import install_codex
install_gemini() # For Gemini CLI
install_codex() # For Codex CLI
Cross-Repo Namespaces
ContextFS automatically detects your git repository and isolates memories:
# In repo A
ctx = ContextFS() # namespace = "repo-<hash-of-repo-a>"
ctx.save("Repo A specific fact")
# In repo B
ctx = ContextFS() # namespace = "repo-<hash-of-repo-b>"
# Won't see Repo A's memories
# Global namespace (shared across repos)
ctx = ContextFS(namespace_id="global")
ctx.save("Shared across all repos")
Configuration
Environment variables:
CONTEXTFS_DATA_DIR=~/.contextfs
CONTEXTFS_EMBEDDING_MODEL=all-MiniLM-L6-v2
CONTEXTFS_CHUNK_SIZE=1000
CONTEXTFS_DEFAULT_SEARCH_LIMIT=10
CONTEXTFS_AUTO_SAVE_SESSIONS=true
CONTEXTFS_AUTO_LOAD_ON_STARTUP=true
Supported Languages
ContextFS supports 50+ file types including Python, JavaScript, TypeScript, Go, Rust, Java, C++, and more. See full list in docs.
Developer Memory Workflow (DMW)
ContextFS enables persistent developer memory across sessions with typed memories:
| Type | Use Case |
|---|---|
fact |
Project configurations, conventions |
decision |
Architectural choices with rationale |
code |
Algorithms, patterns, important snippets |
error |
Bug fixes, error patterns, solutions |
procedural |
Setup guides, deployment steps |
episodic |
Session transcripts, conversations |
See the full Developer Memory Workflow Guide for patterns and examples.
Memory Lineage & Graph Operations
ContextFS tracks memory evolution and relationships with graph-backed lineage:
# Evolve memory (update with history tracking)
contextfs evolve <id> "Updated content" --summary "Why it changed"
# View lineage (ancestors/descendants)
contextfs lineage <id> --direction both
# Merge multiple memories
contextfs merge <id1> <id2> --summary "Combined knowledge" --strategy union
# Split memory into parts
contextfs split <id> "Part 1" "Part 2" --summaries "First|Second"
# Link related memories
contextfs link <id1> <id2> references --bidirectional
# Find connected memories
contextfs related <id> --depth 2
MCP Tools for Graph Operations:
| Tool | Description |
|---|---|
contextfs_evolve |
Update memory with history tracking |
contextfs_merge |
Combine multiple memories into one |
contextfs_split |
Divide memory into separate parts |
contextfs_link |
Create relationships between memories |
contextfs_related |
Find connected memories via graph traversal |
contextfs_lineage |
View memory evolution history |
Relationship Types: references, depends_on, contradicts, supports, supersedes, related_to, derived_from, part_of, implements
Session Management
# List sessions
contextfs sessions
# Save current session
contextfs save --save-session current --label "feature-auth"
# Load session context
contextfs load-session <session_id>
Source tool auto-detected (claude-code, claude-desktop) or set via CONTEXTFS_SOURCE_TOOL.
Web UI
Start the web server to browse and search memories:
contextfs web
# Opens at http://localhost:8000
contextfs web --port 3000 # Custom port
Features:
- Browse all memories with filtering by type, repo, and project
- Side-by-side FTS vs RAG search comparison
- Session browser and message viewer
- Real-time memory statistics
Architecture
┌──────────────────────────────────────────────────────────────────┐
│ ContextFS Core │
├──────────────────────────────────────────────────────────────────┤
│ ┌───────┐ ┌───────┐ ┌───────┐ ┌───────┐ │
│ │ CLI │ │ MCP │ │ Web UI│ │Python │ │
│ │ │ │Server │ │ │ │ API │ │
│ └───┬───┘ └───┬───┘ └───┬───┘ └───┬───┘ │
│ └───────────┴─────┬─────┴───────────┘ │
│ │ │
│ ┌───────▼───────┐ │
│ │ ContextFS() │ │
│ │ core.py │ │
│ └───────┬───────┘ │
│ │ │
│ ┌───────────────┼───────────────┐ │
│ │ │ │ │
│ ┌───────▼───────┐ ┌─────▼─────┐ ┌───────▼───────┐ │
│ │MemoryLineage │ │StorageRouter│ │ AutoIndexer │ │
│ │(Graph Ops) │ │ │ │ (Code Index) │ │
│ └───────┬───────┘ └─────┬─────┘ └───────────────┘ │
│ │ │ │
│ │ ┌───────▼───────┐ │
│ │ │TypedStorage │ ← EdgeRelation, MemoryEdge │
│ │ │ Protocol │ GraphPath, GraphTraversal │
│ │ └───────┬───────┘ │
│ │ │ │
│ └───────┬───────┼───────┬───────────────┐ │
│ │ │ │ │ │
│ ┌───────▼──┐ ┌──▼───┐ ┌─▼────────┐ ┌────▼─────┐ │
│ │ SQLite │ │Chroma│ │PostgreSQL│ │ FalkorDB │ │
│ │ + FTS5 │ │ DB │ │ +pgvector│ │ (Cypher) │ │
│ │(default) │ │(RAG) │ │ (hosted) │ │ (graph) │ │
│ └──────────┘ └──────┘ └──────────┘ └──────────┘ │
└──────────────────────────────────────────────────────────────────┘
Storage Backends
| Backend | Purpose |
|---|---|
| SQLite + FTS5 | Default local storage, keyword search, sessions |
| ChromaDB | Vector embeddings, semantic/RAG search |
| PostgreSQL + pgvector | Hosted deployments, team sharing |
| FalkorDB | Advanced graph queries via Cypher |
Typed Storage Protocol
The StorageProtocol provides a unified interface across backends with typed models:
EdgeRelation- 20+ relationship types (references, depends_on, contradicts, etc.)MemoryEdge- Typed edges with weights and metadataGraphPath/GraphTraversal- Path finding and subgraph queries
License
MIT
Authors
Matthew Long and The YonedaAI Collaboration
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
E2B
Using MCP to run code via e2b.