MCP Server

Connect AI agents to your infrastructure via the Model Context Protocol. Query resources, relationships, and context packs programmatically.

Overview

Knowledge Tree includes a Model Context Protocol (MCP)server that allows AI agents (like Claude, GPT, or custom agents) to query your infrastructure directly. The MCP server exposes tools for searching resources, traversing relationships, and assembling context packs.

Transports

The MCP server supports two transport modes:

Transport	Use Case
stdio	Claude Desktop, local agents
HTTP+SSE	Remote agents, web applications

Available Tools

The MCP server exposes these tools to connected agents:

Tool	Description	Parameters
`search_resources`	Search for resources	query, type, provider, limit
`get_resource`	Get a specific resource	type, id
`get_relationships`	Get edges for a resource	id, direction, edgeType
`traverse`	Graph traversal	startId, direction, edgeTypes, maxDepth
`assemble_context`	Build a context pack	packType, target, maxTokens
`query_graph`	Execute a Cypher query	cypher, params

Context Packs

Context packs are pre-assembled bundles of relevant infrastructure information, sized to fit within LLM token budgets. Available pack types:

Pack Type	Purpose
`runbook`	Operational runbook for a resource
`architecture`	Architecture description with neighbors
`impact`	Impact analysis (what depends on this)
`inventory`	Resource inventory matching a query
`search`	Semantic search across documentation

Configuration

# In your MCP client config (e.g., Claude Desktop)
{
  "mcpServers": {
    "knowledge-tree": {
      "command": "kt-agent",
      "args": ["mcp", "--config", "configs/dev.yaml"],
      "transport": "stdio"
    }
  }
}

HTTP+SSE Transport

For remote agents, the MCP server also listens on the configured MCP port (default 9090). Point your agent's MCP client athttp://your-server:9090/mcp.