---

name: mcp-servers description: Use and develop MCP servers for agent capabilities. Includes Temporal, Qdrant, Memory, Discord, and Registry MCP servers.

MCP Servers

Kubani provides several MCP servers that extend agent capabilities.

Available MCP Servers

Quick Start

Installation

Copy# Install all MCP servers
uv pip install -e kubani/mcp/servers/temporal
uv pip install -e kubani/mcp/servers/qdrant
uv pip install -e kubani/mcp/servers/memory
uv pip install -e kubani/mcp/servers/discord

Claude Code Configuration

Add to .claude/mcp.json:

Copy{
  "mcpServers": {
    "temporal": {
      "command": "temporal-mcp",
      "env": {
        "TEMPORAL_HOST": "temporal.almckay.io:7233",
        "TEMPORAL_NAMESPACE": "kubani"
      }
    },
    "qdrant": {
      "command": "qdrant-mcp",
      "env": {
        "QDRANT_HOST": "qdrant.almckay.io",
        "QDRANT_PORT": "6333"
      }
    },
    "memory": {
      "command": "memory-mcp",
      "env": {
        "QDRANT_HOST": "qdrant.almckay.io",
        "NEO4J_URI": "bolt://neo4j.almckay.io:7687",
        "REDIS_HOST": "redis.almckay.io"
      }
    },
    "discord": {
      "command": "discord-mcp",
      "env": {
        "DISCORD_BOT_TOKEN": "${DISCORD_BOT_TOKEN}"
      }
    }
  }
}

Temporal MCP Server

Manage Temporal workflows and schedules.

Tools

• list_workflows - List workflows with filtering
• get_workflow - Get workflow details
• start_workflow - Start a new workflow
• signal_workflow - Send signals to workflows
• query_workflow - Query workflow state
• cancel_workflow - Cancel a workflow
• terminate_workflow - Force terminate
• list_schedules - List schedules
• pause_schedule / unpause_schedule - Manage schedules
• trigger_schedule - Trigger immediate execution

Example Usage

CopyList all running k8s-monitor workflows

CopyStart a new news-digest workflow with ID "digest-2024-01-11"

CopyGet the history of workflow "failed-workflow-123" to debug the failure

Qdrant MCP Server

Vector database operations for semantic search.

Tools

• list_collections - List all collections
• create_collection - Create a new collection
• delete_collection - Delete a collection
• upsert_vectors - Insert/update vectors
• search_vectors - Semantic similarity search
• get_point - Get a specific point
• delete_points - Delete points
• scroll_points - Paginate through points
• count_points - Count points

Example Usage

CopyCreate a collection called "agent-learnings" with 1536 dimensions

CopySearch for vectors similar to "kubernetes memory issues" in the learnings collection

Memory MCP Server

Unified memory system combining Qdrant, Neo4j, and Redis.

Tools

Learnings:

• store_learning - Store agent learnings
• query_learnings - Semantic search learnings
• get_agent_learnings - Get agent's learnings

Knowledge:

• store_knowledge - Store domain knowledge
• query_knowledge - Search knowledge
• get_knowledge_graph - Explore relationships
• find_related_topics - Find related topics

Relationships:

• create_relationship - Create entity relationships
• get_entity_relationships - Get relationships

Cache:

• cache_set / cache_get / cache_delete - Fast caching

Utilities:

• get_memory_stats - Memory system stats
• consolidate_learnings - Consolidate similar learnings

Example Usage

CopyStore a learning from k8s-monitor: "OOM kills indicate memory pressure"
with confidence 0.85 and tags ["kubernetes", "memory"]

CopyQuery learnings about "pod restart issues" with minimum confidence 0.7

CopyShow the knowledge graph around "kubernetes/memory-management"

Discord MCP Server

Discord integration for notifications and approvals.

Tools

• send_message - Send a message to a channel
• send_embed - Send rich embed message
• add_reaction - Add reaction to message
• get_reactions - Get message reactions
• create_thread - Create a thread
• wait_for_reaction - Wait for specific reaction

Example Usage

CopySend an alert to the alerts channel about a pod failure

CopyPost a skill proposal to the learning channel and wait for approval reactions

Developing New MCP Servers

Template

Copy"""
My MCP Server implementation.
"""

import os
from mcp.server.fastmcp import FastMCP

def create_server() -> FastMCP:
    mcp = FastMCP(
        name="My MCP Server",
        instructions="Description of what this server does.",
    )

    @mcp.tool()
    async def my_tool(
        param1: str,
        param2: int = 10,
    ) -> dict:
        """
        Tool description.

        Args:
            param1: Description of param1
            param2: Description of param2 (default: 10)

        Returns:
            Result description
        """
        # Implementation
        return {"result": "value"}

    return mcp


async def main():
    server = create_server()
    
    transport = os.environ.get("MCP_TRANSPORT", "stdio")
    if transport == "stdio":
        from mcp.server.stdio import stdio_server
        async with stdio_server() as (read, write):
            await server.run(read, write, server.create_initialization_options())
    elif transport == "sse":
        await server.run_sse_async()


if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

Project Structure

Copytools/my-mcp-server/
├── src/my_mcp/
│   ├── __init__.py
│   ├── server.py      # MCP server implementation
│   └── models.py      # Pydantic models
├── tests/
│   ├── __init__.py
│   ├── conftest.py
│   └── test_server.py
├── pyproject.toml
└── README.md

Testing

Copy# Run tests
cd tools/my-mcp-server
pytest

# Test manually
python -m my_mcp.server

Registration

Add to the registry for discovery:

Copykubani-dev mcp register my-mcp-server

Best Practices

1. Use descriptive tool names that indicate the action
2. Provide detailed docstrings for tool discovery
3. Use Pydantic models for complex inputs/outputs
4. Handle errors gracefully with informative messages
5. Add comprehensive tests for all tools
6. Document environment variables in README
7. Support both stdio and SSE transports