OpenClaude MCP Server
Provides tools for managing local Ollama models and sessions, including starting, prompting, and stopping sessions, model inventory, and health checks.
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., "@OpenClaude MCP Serverstart a session with llama3 and ask it to write a short story"
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.
OpenClaude MCP Server
📖 Installation Guide — quick start, manual setup, and troubleshooting
A high-performance control plane for Ollama-based local LLM sessions, with background memory consolidation (KAIROS), hybrid cloud planning (ULTRAPLAN), and real-time fleet monitoring. Optimized for RTX 4090 environments.
Features
Background Memory (KAIROS): Auto-consolidates session data into
MEMORY.mdusing asynchronous background cycles. Configurable poll interval and consolidation budget. State persists across restarts.Session Management: Launch, prompt, monitor, and stop multiple OpenClaude subprocesses concurrently. Persistent across restarts with usage analytics (prompts, output chars, estimated tokens).
Multimodal Input: Send images (png, jpeg, webp, gif) alongside text instructions via the
send_multimodaltool. Works with any vision-capable Ollama model.Ollama Control Plane: Model management with VRAM/speed/context metadata, health checks, default persistence.
Hybrid Planning (ULTRAPLAN): Optional cloud Opus/Sonnet handoff for complex reasoning, local execution. Token cost tracking included.
Fleet Observability: React dashboard with real-time SSE push, xterm.js terminal, system log viewer, KAIROS status, interactive API playground.
Interactive Examples: Live "Run" buttons on every example — calls the real backend, shows response JSON with timing.
API Reference: Searchable documentation with parameter tables, return schemas, env vars, error codes for all 15 tools.
Safety Guardrails: Kid-safe mode with content filters, proactive privacy reminders, caregiver alerts (file + webhook).
Dual Transport: MCP SSE (Claude Desktop) + REST bridge (webapp/curl) on the same port.
Quick Start
git clone https://github.com/sandraschi/openclaude-mcp
cd openclaude-mcp
justThis opens an interactive dashboard showing all available commands. Run just bootstrap to install dependencies, then just serve or just dev to start.
Manual Setup
If you don't have just installed:
.\setup.ps1
.\start.ps1
SSE Endpoint:
http://localhost:10932/sseFleet Dashboard:
http://localhost:10933API Events:
http://localhost:10932/api/events(SSE push)Health API:
http://localhost:10932/api/healthOr with Docker: docker compose up
Available Tools (15)
Tool | Action |
| Initialize a new Ollama session |
| Execute a prompt in an active session |
| Send text + images to a session (png, jpeg, webp, gif) |
| Get output/status/usage/elapsed of a session |
| List all active sessions |
| Stop and clean up a session |
| Activate background memory consolidation |
| Halt background memory consolidation |
| Retrieve KAIROS consolidation log |
| Inventory of available Ollama models |
| Set default model for new sessions |
| Check VRAM and load status |
| Hybrid cloud/local planning cycle |
| Prefab UI fleet status overview |
| Raw fleet status JSON |
| [KID-SAFE] Notify caregivers of high-risk interaction |
Architecture
┌─────────────────┐
│ Claude Desktop │── SSE ──────┐
│ Cursor / VSCode │ │
└─────────────────┘ │
▼
┌──────────────┐ REST ┌────────────────────────────┐ stdio (NDJSON) ┌──────────────────┐
│ React Webapp│───────►│ server.py (FastMCP 3.2) │◄────────────────│ OpenClaude CLI │
│ (SSE push) │◄───────│ + SessionStore │ subprocess │ (Node.js/Bun) │
└──────────────┘ │ + ModelRouter │◄───────────────►│ + Ollama :11434 │
│ + KairosController │ └──────────────────┘
│ + SessionPersistence │
│ + SSE Event Bus │
└────────────────────────────┘Configuration (env vars)
Variable | Default | Description |
|
| Backend port |
| — | REST auth token (disable if unset) |
|
| OpenClaude source path |
|
| Anthropic model for ULTRAPLAN |
|
| Persistence directory |
|
| KAIROS daemon poll interval |
|
| Max consolidations per session |
| — | Webhook for caregiver alerts |
Prerequisites
Ollama (running locally)
Node.js (v20+)
Python 3.13+ with
uvLocal model pulled (e.g.,
gemma4:26b,qwen3.5:35b-a3b)
Test Suite
uv run pytest # 79 tests (unit + smoke + integration + e2e)
uv run pytest tests/unit/ # fast, no external deps
uv run pytest tests/e2e/ # Anthropic API mocked via respxLicense
MIT
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/sandraschi/openclaude-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server