MCP Hub
Back to servers

CPersona

Persistent AI memory server with 3-layer hybrid search (vector + FTS5 + keyword), confidence scoring via Reciprocal Rank Fusion, episodic/profile memory, and 16 tools. Zero LLM dependency. Works standalone with Claude Desktop and Claude Code. MIT licensed.

glama
Stars
2
Updated
Apr 5, 2026
Validated
Apr 6, 2026

cloto-mcp-servers

License Tests

MCP server collection for the ClotoCore platform.

CPersona — AI Memory Server

CPersona (servers/cpersona/) is an MCP Memory Server that gives Claude persistent memory. See the cpersona README for full documentation.

Standalone users: For use with Claude Desktop, Claude Code, or other MCP clients without ClotoCore, see the cpersona standalone repository.

  • 3-layer hybrid search (vector + FTS5 full-text + keyword) with RRF merge
  • Confidence scoring with dynamic time decay, recall boost, and completion factor
  • Episodic memory (conversation summarization) and profile memory (user attributes)
  • Zero LLM dependency — pure data server, all intelligence stays in the calling agent
  • 16 tools including health check, auto-calibration, JSONL export/import, and agent merge
  • Agent namespace isolation
  • stdio + Streamable HTTP transport
  • Single-file SQLite DB (schema v7, auto-migrating)

License: MIT — free to use from any MCP host.

Quick Start (Claude Desktop / Claude Code)

git clone https://github.com/Cloto-dev/cloto-mcp-servers.git
cd cloto-mcp-servers/servers
python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS / Linux
# source .venv/bin/activate

pip install -r requirements.lock

Claude Desktop — add to claude_desktop_config.json:

{
  "mcpServers": {
    "embedding": {
      "command": "/path/to/servers/.venv/bin/python",
      "args": ["/path/to/servers/embedding/server.py"],
      "env": {
        "EMBEDDING_PROVIDER": "onnx_jina_v5_nano",
        "EMBEDDING_HTTP_PORT": "8401"
      }
    },
    "cpersona": {
      "command": "/path/to/servers/.venv/bin/python",
      "args": ["/path/to/servers/cpersona/server.py"],
      "env": {
        "CPERSONA_DB_PATH": "/home/yourname/.claude/cpersona.db",
        "CPERSONA_EMBEDDING_MODE": "http",
        "CPERSONA_EMBEDDING_URL": "http://127.0.0.1:8401/embed",
        "CPERSONA_VECTOR_SEARCH_MODE": "remote"
      }
    }
  }
}

Windows: use .venv/Scripts/python.exe instead of .venv/bin/python, and C:/Users/yourname/.claude/cpersona.db for the DB path.

Note: Claude Desktop on Windows has known issues with stdio MCP transport (payload size limits, silent drops). If you experience problems, consider using cpersona's Streamable HTTP transport instead. See the Zenn Book Ch.2 for details.

Claude Code:

# macOS / Linux
claude mcp add-json embedding '{
  "type": "stdio",
  "command": "/path/to/servers/.venv/bin/python",
  "args": ["/path/to/servers/embedding/server.py"],
  "env": {
    "EMBEDDING_PROVIDER": "onnx_jina_v5_nano",
    "EMBEDDING_HTTP_PORT": "8401"
  }
}' -s user

claude mcp add-json cpersona '{
  "type": "stdio",
  "command": "/path/to/servers/.venv/bin/python",
  "args": ["/path/to/servers/cpersona/server.py"],
  "env": {
    "CPERSONA_DB_PATH": "/home/yourname/.claude/cpersona.db",
    "CPERSONA_EMBEDDING_MODE": "http",
    "CPERSONA_EMBEDDING_URL": "http://127.0.0.1:8401/embed",
    "CPERSONA_VECTOR_SEARCH_MODE": "remote"
  }
}' -s user

Adjust paths to match your environment. For detailed setup instructions, see the Zenn Book Ch.2.

All Servers

IDTypeDescription
memory.cpersonaMemoryCPersona persistent memory with FTS5 search (MIT)
tool.embeddingToolVector embedding generation (ONNX + API providers)
tool.terminalToolSandboxed command execution
tool.agent_utilsToolDeterministic utilities (time, math, UUID, units, encode/decode)
tool.cronToolCRON job management
tool.websearchToolWeb search integration (SearXNG, Tavily, DuckDuckGo)
tool.researchToolResearch synthesis with search and LLM delegation
tool.imagegenToolStable Diffusion image generation
mind.cerebrasMindCerebras ultra-high-speed reasoning
mind.deepseekMindDeepSeek reasoning engine
mind.claudeMindClaude/Anthropic reasoning engine
mind.ollamaMindOllama local LLM
vision.gaze_webcamVisionWebcam gaze tracking
vision.captureVisionImage capture with Ollama + OCR
voice.sttVoiceSpeech-to-text (Whisper)
output.avatarOutputVRM expression, idle behavior, and VOICEVOX TTS (Rust)
io.discordI/OBidirectional Discord communication via MGP events (Rust)
tool.exampleExampleReference MGP server implementation for new authors

MGP Protocol Documentation

ClotoCore extends MCP with MGP (Multi-Agent Gateway Protocol) — a strict superset adding security, lifecycle, and bidirectional communication.

Build Your Own Server

See servers/example/server.py for a complete reference implementation, or follow the Quickstart Guide to create a working MCP/MGP server in 5 minutes.

Setup (all servers)

cd servers
python -m venv .venv

# Activate (Windows Git Bash)
source .venv/Scripts/activate
# Activate (Linux/macOS)
# source .venv/bin/activate

pip install --upgrade pip
for d in */; do [ -f "$d/pyproject.toml" ] && pip install "$d"; done

Or use the setup script:

bash scripts/setup-deps.sh

Integration with ClotoCore

This repo is consumed by ClotoCore via mcp.toml's [paths] section:

[paths]
servers = "/path/to/cloto-mcp-servers/servers"

Server args use ${servers} variable expansion:

[[servers]]
id = "tool.terminal"
command = "python"
args = ["${servers}/terminal/server.py"]

Testing

cd servers
python -m pytest tests/ -v

License

This repository uses a dual-license model:

ComponentLicense
CPersona (servers/cpersona/)MIT
MGP Protocol (docs/MGP_*.md)MIT
All other servers and codeBSL 1.1 (converts to MIT on 2028-02-14)

See LICENSE for the BSL 1.1 terms. CPersona and MGP are independently MIT-licensed to enable adoption by any MCP host without restriction.

Contact: ClotoCore@proton.me

Reviews

No reviews yet

Sign in to write a review