Sekha MCP
Universal AI Memory Controller - Persistent, Sovereign, Open
Installation
npx sekha-mcpAsk AI about Sekha MCP
Powered by Claude Β· Grounded in docs
I know everything about Sekha MCP. Ask me about installation, configuration, usage, or troubleshooting.
0/500
Reviews
Documentation
Sekha MCP Server
Model Context Protocol Server for Sekha Memory
π v0.2.0 Release - Multi-Provider Support
Sekha MCP v0.2.0 is now compatible with the new Sekha v0.2.0 multi-provider architecture!
What's New:
- β Works with Sekha v0.2.0 controller's multi-provider routing
- β Automatic provider fallback (Ollama, OpenAI, Anthropic, etc.)
- β Vision support (GPT-4o, Kimi 2.5) - just include images!
- β Cost-aware model selection
- β Multi-dimensional embeddings (per-dimension ChromaDB collections)
- β Claude Desktop & Claude Code support - memory in both apps!
- β No API changes - fully backward compatible!
What is Sekha MCP?
MCP (Model Context Protocol) server that exposes Sekha memory tools to any MCP-compatible client:
- β Claude Desktop - Anthropic's desktop app
- β Claude Code - VS Code extension (works with Ollama, Anthropic, or any provider)
- β Any MCP client - Standard protocol implementation
Supported Tools:
- β
memory_store- Save conversations - β
memory_search- Semantic search - β
memory_get_context- Retrieve relevant context - β
memory_update- Update conversation metadata - β
memory_prune- Get cleanup recommendations - β
memory_export- Export your data - β
memory_stats- View usage statistics
Total: 7 MCP tools
π Documentation
Complete guide: docs.sekha.dev/integrations/mcp
π Quick Start
1. Install Sekha
# Deploy Sekha v0.2.0 stack with multi-provider support
git clone https://github.com/sekha-ai/sekha-docker.git
cd sekha-docker
docker compose -f docker/docker-compose.prod.yml up -d
2. Configure Your MCP Client
Option A: Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
{
"mcpServers": {
"sekha": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"--network=host",
"ghcr.io/sekha-ai/sekha-mcp:v0.2.0"
],
"env": {
"CONTROLLER_URL": "http://localhost:8080",
"CONTROLLER_API_KEY": "your-mcp-api-key-here"
}
}
}
}
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json
Option B: Claude Code (VS Code Extension)
Add to VS Code settings.json or workspace config:
{
"mcpServers": {
"sekha": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"--network=host",
"ghcr.io/sekha-ai/sekha-mcp:v0.2.0"
],
"env": {
"CONTROLLER_URL": "http://localhost:8080",
"CONTROLLER_API_KEY": "your-mcp-api-key-here"
}
}
}
}
Claude Code Configuration:
Claude Code lets you choose your LLM provider separately from memory:
{
// Sekha provides memory (via MCP)
"mcpServers": {
"sekha": { /* config above */ }
},
// Configure your LLM provider (Claude Code supports multiple)
"claudeCode.apiProvider": "ollama", // or "anthropic"
"claudeCode.ollamaUrl": "http://localhost:11434",
"claudeCode.ollamaModel": "llama3.1:8b"
}
This means:
- Use Ollama (or other LLM) locally for generation (fast, private, free)
- Use Sekha MCP for memory (persistent across sessions)
- Best of both worlds!
3. Restart Your Client
- Claude Desktop: Restart the app
- Claude Code: Reload VS Code window (
Cmd+Shift+Pβ "Reload Window")
Sekha memory tools will now appear!
See setup guides:
π― Use Cases
Claude Desktop - Interactive Conversations
- Full-featured desktop app with Sekha memory
- Perfect for brainstorming, research, general chat
- Uses Anthropic's Claude models
Claude Code - Development Workflow Examples
- VS Code extension with code-aware features
- Use with Ollama for fast, local, private coding
- Or use with Anthropic/OpenAI for powerful cloud models
- Sekha memory works with any provider you configure
API Integration - Programmatic Access
- Use sekha-proxy for OpenAI-compatible API
- Multi-provider routing via LLM bridge
- Same memory as Claude apps
π§ Development
# Clone
git clone https://github.com/sekha-ai/sekha-mcp.git
cd sekha-mcp
# Install
pip install -e .
# Run locally
python -m sekha_mcp
# Test
pytest
π MCP Tools Reference
memory_store
Store a conversation in Sekha.
Parameters:
label(string) - Conversation labelmessages(array) - Message array (supports images in v0.2.0!)folder(string, optional) - Organization folderimportance(int, optional) - 1-10 scale
memory_search
Search conversations semantically.
Parameters:
query(string) - Search querylimit(int) - Max resultsfolder(string, optional) - Search within folder
memory_get_context
Assemble optimal context for LLM.
Parameters:
query(string) - Context querycontext_budget(int) - Token limitfolders(array, optional) - Limit to specific folders
memory_update
Update conversation metadata.
Parameters:
conversation_id(string) - Conversation UUIDlabel(string, optional) - New labelfolder(string, optional) - New folderimportance(int, optional) - New importance (1-10)status(string, optional) - active/archived
memory_prune
Get cleanup recommendations.
Parameters:
min_age_days(int, optional) - Minimum agemax_importance(int, optional) - Max importance to considerlimit(int, optional) - Max suggestions
memory_export
Export conversations.
Parameters:
format(string) - json or markdownfolder(string, optional) - Export specific folder
memory_stats
Get memory usage statistics.
Parameters: None
Returns:
- Total conversations
- Total messages
- Storage usage
- Folder breakdown
- Provider stats (v0.2.0) - which models are being used
ποΈ Architecture
βββββββββββββββββββββββββββββββββββββββββββββββ
β MCP Clients β
β - Claude Desktop (Anthropic) β
β - Claude Code (Ollama/Anthropic/etc.) β
β - Any MCP-compatible client β
ββββββββββββββββββ¬βββββββββββββββββββββββββββββ
β
βΌ
ββββββββββββββββββ
β Sekha MCP β β This repository
β Server β
ββββββββββ¬ββββββββ
β
βΌ
ββββββββββββββββββ
β Controller β β Memory APIs
β (Rust) β
ββββββββββ¬ββββββββ
β
βΌ
ββββββββββββββββββ
β ChromaDB β β Vector storage
β Redis β β Cache
ββββββββββββββββββ
Separate from LLM routing:
βββββββββββββββββββββββββββββββββββββββββββββββ
β API Clients β Proxy β Bridge β Providers β
β (OpenAI SDK compatible) β
βββββββββββββββββββββββββββββββββββββββββββββββ
Key Points:
- MCP provides memory tools only
- Claude Desktop/Code handle their own LLM connections
- Controller stores all conversations regardless of source
- Same memory accessible from Claude apps and API
π Links
- Main Repo: sekha-controller
- Proxy (API): sekha-proxy
- Docker Deploy: sekha-docker
- Docs: docs.sekha.dev
- Website: sekha.dev
- Discord: discord.gg/sekha
π Changelog
See CHANGELOG.md for full release history.
π License
AGPL-3.0 - License Details
