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., "@Nouva MCP Server search memory for notes from yesterday"
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.
Nouva MCP (Model Context Protocol) Server is a centralized repository for modular, portable, and detachable Personalized Skills and Memory. It is written in Python (FastMCP) and runs on Gading's agent-host. You can run it locally or in a containerized environment.
The server supports dual transport:
Stdio Transport: Used natively by OpenClaw on the local machine.
SSE (Server-Sent Events) Transport: Runs an HTTP server on port 8000 to be accessed by external clients like Cursor, Windsurf, or Hermes.
[AI Agent / IDE Client]
(OpenClaw, Cursor, Zed, Claude Code)
ā
ā¼ (via stdio / SSE)
āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā
ā Nouva MCP Server ā
ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā
ā ā Skills Engine ā ā (dynamic loader under src/skills/)
ā ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā ā
ā ā ā Memory Engine ā ā ā (pgvector recall + SQL analytics)
ā ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā ā
ā ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā ā
ā ā ā Persona Engine ā ā ā (startup persona bootstrap from markdown folders)
ā ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā ā
ā ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā ā
ā ā ā MCP Management ā ā ā (scaffolding & creating new skills)
ā ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā ā
ā ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā ā
ā ā ā Other Skills ā ā ā (custom tools, morning-report, etc.)
ā ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā ā
ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā
āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāPython-based: tools should be implemented as native Python inside this repository.
Self-contained: avoid wrappers that call scripts outside the repo.
Secrets: keep secrets out of git and out of the repo by default. Prefer environment variables or host-mounted secret files.
Related MCP server: Total Recall
nouva-mcp-server/
āāā src/
ā āāā main.py # Entrypoint & Dynamic Skill Loader
ā āāā skills/ # Modular skill directories (native Python)
ā ā āāā memory_engine/
ā ā ā āāā README.md # Setup and operational notes for memory_engine
ā ā ā āāā SKILL.md # Agent routing guidance for memory tools
ā ā ā āāā memory_config.example.json # Template config for memory variables
ā ā ā āāā memory_scripts/
ā ā ā ā āāā memory_auto_sync.py # Memory auto-sync cron script
ā ā ā ā āāā memory_query_context.py # Hybrid search query engine
ā ā ā ā āāā memory_query_analyze.py # Deterministic structured analytics executor
ā ā ā ā āāā memory_db/ # DB helpers and init scripts
ā ā ā ā ā āāā memory_init_db.py
ā ā ā ā ā āāā memory_db_helper.py
ā ā ā ā ā āāā memory_analytics_repo.py
ā ā ā ā āāā memory_sync/
ā ā ā ā āāā memory_summary_sync.py
ā ā ā ā āāā memory_sync.py
ā ā ā ā āāā memory_analytics_sync.py
ā ā ā āāā tools/
ā ā ā āāā memory_query.py # Tool: memory_query
ā ā ā āāā memory_analyze.py # Tool: memory_analyze
ā ā ā āāā memory_grep.py # Tool: memory_grep
ā ā ā āāā memory_read_file.py # Tool: memory_read_file
ā ā ā āāā session_write.py # Tool: session_write (batch transcript writer)
ā ā āāā persona_engine/
ā ā ā āāā README.md # Persona setup and startup prompt integration
ā ā ā āāā SKILL.md # Agent guidance for persona loading
ā ā ā āāā scripts/
ā ā ā ā āāā util/
ā ā ā ā āāā persona_loader.py # Persona validation and prompt assembly
ā ā ā āāā tools/
ā ā ā āāā persona_list.py # Tool: persona_list
ā ā ā āāā persona_get_prompt.py # Tool: persona_get_prompt
ā ā āāā mcp_management/
ā ā ā āāā SKILL.md # Scaffolding guidelines (Resource)
ā ā ā āāā tools/
ā ā ā āāā skill_create.py # Tool: skill_create
ā āāā utils/
āāā personas/ # Private persona folders + public example template
āāā requirements.txt # Python dependencies
āāā ROADMAP_PLAN.md # Long-term development planYou can run Nouva MCP Server either natively using Python or containerized via Docker.
Install minimal system dependencies required for python packages compilation (e.g. psycopg2 for PostgreSQL):
apt update && apt install -y git build-essential libpq-dev python3-devClone the repository and install the dependencies:
git clone https://github.com/nouverse-tech/nouva-mcp-server.git
cd nouva-mcp-server
pip install -r requirements.txt --break-system-packagesCopy the memory configuration file:
# Setup memory config (includes database credentials)
cp src/skills/memory_engine/memory_config.example.json src/skills/memory_engine/memory_config.json
# Edit src/skills/memory_engine/memory_config.json with your local details:
# - database.host/port/name/user/password (or database.url for full connection string)
# - embedding.url and embedding.model
# - llm.url and llm.model
# - memory_paths.active_memory_dirpython3 src/main.py --transport stdioTo enable one persona automatically for every new chat session:
python3 src/main.py --transport stdio --default-persona=nouva-exampleIf --default-persona is omitted, persona mode remains off and no startup persona prompt is injected by default.
We provide a lightweight setup to build and run the MCP server in a Docker container.
docker build -t nouverse/nouva-mcp-server:latest .Ensure you have mapped your config files inside docker-compose.yml:
docker compose up -dWe provide a development configuration that spins up both the PostgreSQL (with pgvector extension) database and the MCP Server inside a single docker network.
Copy the configuration file:
cp src/skills/memory_engine/memory_config.example.json src/skills/memory_engine/memory_config.json(Note: The default credentials in memory_config.example.json are pre-configured to match the local PostgreSQL container in docker-compose.dev.yml)
Run the development docker compose:
docker compose -f docker-compose.dev.yml up -dInitialize the development database: Run the initialization script inside the container (or locally if you have python dependencies installed):
docker exec -it nouva-mcp-server-dev python3 src/skills/memory_engine/memory_scripts/memory_db/memory_init_db.pyAdd the following configuration to your OpenClaw MCP config (path depends on your installation):
{
"mcpServers": {
"nouva-mcp": {
"command": "python3",
"args": [
"/absolute/path/to/nouva-mcp-server/src/main.py",
"--transport",
"stdio"
]
}
}
}Cursor reads MCP server configs from:
Project-scoped: .cursor/mcp.json
User-scoped: ~/.cursor/mcp.json
Option A ā Connect via SSE (recommended if the server runs separately, e.g. Docker):
Start the server in SSE mode on the host:
python3 src/main.py --transport sse --port 8000Add this to .cursor/mcp.json:
{
"mcpServers": {
"nouva-mcp": {
"url": "http://<host>:8000/sse"
}
}
}Option B ā Run via stdio (Cursor spawns the process locally):
{
"mcpServers": {
"nouva-mcp": {
"command": "python3",
"args": [
"/absolute/path/to/nouva-mcp-server/src/main.py",
"--transport",
"stdio"
]
}
}
}Cursor MCP docs: https://cursor.com/docs/mcp
Zed stores MCP server configs in its settings file under context_servers. You can add this via UI ("Add Context Server") or edit settings manually.
Manual configuration example (~/.config/zed/settings.json), using stdio:
{
"context_servers": {
"nouva-mcp": {
"command": "python3",
"args": [
"/absolute/path/to/nouva-mcp-server/src/main.py",
"--transport",
"stdio"
],
"env": {}
}
}
}Zed MCP docs: https://zed.dev/docs/ai/mcp.html
Start the server in SSE mode:
python3 src/main.py --transport sse --port 8000Then connect using:
URL: http://<host>:8000/sse
Transport: SSE
Connecting to the MCP server is enough for tool discovery. For best results, add a short routing rule in your agent instructions:
Use memory_analyze for aggregation/time-series questions, but call it with structured arguments only after the agent parses the user's natural-language request. The analytics contract supports date lists, top values, weekday distributions, distinct-date counts, counts by period, grouped top values, and average importance.
Use memory_query for detailed recall and context.
Use persona_get_prompt only at new-session bootstrap time when a persona is explicitly selected or a default persona is configured.
This project is licensed under the GNU Affero General Public License v3 (AGPL-3.0). See the LICENSE file for the full license text.
For detailed setup instructions regarding the 2-lane memory architecture, pgvector recall, SQL analytics, embedding settings, database initialization, and memory sync operations, please refer to the Memory Engine README.
For startup persona packs, markdown folder structure, required files (IDENTITY.md, SOUL.md, USER.md), fail-fast validation, and --default-persona setup, please refer to the Persona Engine README.
Since all daily logs and summaries are stored in a clean Markdown format (YYYY-MM-DD.md and _summaries/YYYY-MM-DD.summary.md), you can easily open the active/archived memory directories in Obsidian to explore your memories visually as an interconnected knowledge graph.
This server cannot be installed
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Persistent memory and knowledge management for AI agents with semantic search and 50+ tools.
Secure, user-owned long-term memory for AI agents over OAuth-protected remote MCP. Save, search, recall, update, and govern preferences, project context, decisions, and task state across ChatGPT, Claude, Copilot, IDEs, and CLIs.
Agent-first skill marketplace with USK open standard for Claude, Cursor, Gemini, Codex CLI.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/nouverse-tech/nouva-mcp-server'
If you have feedback or need assistance with the MCP directory API, please join our Discord server