majiayu000 MCP Integration
Guide for developing and using MCP servers in Kubani.
Available MCP Servers
| Server | Purpose | Location |
|---|---|---|
| temporal-mcp | Workflow orchestration | kubani/mcp/servers/temporal/ |
| qdrant-mcp | Vector search | kubani/mcp/servers/qdrant/ |
| memory-mcp | Unified memory (Qdrant + Neo4j + Redis) | kubani/mcp/servers/memory/ |
| discord-mcp | Discord integration | kubani/mcp/servers/discord/ |
| skills-mcp | Skills registry | kubani/mcp/servers/skills/ |
Using MCP in Agents
from kubani.framework.mcp import get_mcp_client
client = get_mcp_client()
# Memory operations
await client.memory.store_learning(agent_id="k8s-monitor", ...)
# Temporal operations
workflows = await client.temporal.list_workflows(status="running")
# Discord operations
await client.discord.send_embed(channel_id=..., title="Alert", ...)
# Qdrant operations
results = await client.qdrant.search_vectors(collection="skills", ...)
Creating a New MCP Server
Template (Python FastMCP)
from mcp.server.fastmcp import FastMCP
mcp = FastMCP("my-mcp-server")
@mcp.tool()
async def my_tool(param: str) -> dict:
"""Tool description for LLM understanding."""
return {"result": param}
if __name__ == "__main__":
mcp.run()
Project Structure
kubani/mcp/servers/my-server/
├── src/my_mcp/
│ ├── __init__.py
│ ├── server.py # MCP server implementation
│ └── models.py # Pydantic models
├── tests/
├── pyproject.toml
└── README.md
Tool Design Best Practices
- Workflow-oriented: Tools should be agent-centric, not thin API wrappers
- Descriptive docstrings: These become the tool descriptions Claude sees
- Error handling: Return structured errors with suggestions, don't raise
- Annotations: Use
readOnlyHint,destructiveHint,idempotentHint - Actionable output: Curate responses — highlight issues, suggest actions
Registration
Add to .claude/mcp.json for Claude Code, or register for cluster use:
kubani mcp register my-mcp-server
Deployment
MCP servers run as cluster services via Kubernetes deployments in infrastructure/gitops/apps/mcp-servers/.