MODEL CONTEXT PROTOCOL

Connect your agents
to structured memory

The MCP server exposes your Konnektr Graph as tools. Claude, Cursor, and custom agents can create entities, build relationships, and query the graph through natural conversation.

Read MCP Docs Deploy a Graph

Connect in minutes

Every Konnektr Graph instance includes a hosted MCP server. Just add the httpUrl to your client config with your resource ID.

Streamable HTTP Transport

Low-latency bidirectional communication via httpUrl.

Resource-Scoped Access

Each graph instance has its own isolated endpoint.

Works with any MCP client

Claude Desktop, Cursor, Windsurf, or custom agents.

claude_desktop_config.jsonJSON

{
  "mcpServers": {
    "konnektr-memory": {
      "httpUrl": "https://mcp.graph.konnektr.io/mcp?resource_id={resource_id}"
    }
  }
}

16 tools for graph operations

Your agent gets a complete toolkit for creating, querying, and managing entities in the knowledge graph.

Search & Query

search_digital_twins

Semantic search across all entities using embeddings.

{ "search_text": "authentication", "limit": 10 }

search_models

Find DTDL models by name or description.

{ "search_text": "user module" }

query_digital_twins

Execute Cypher queries for graph traversal.

{ "query": "MATCH (n:Twin) RETURN n LIMIT 5" }

Create & Update

create_or_replace_digital_twin

Create or update an entity with validated properties.

{ "twin_id": "sensor-1", "model_id": "dtmi:...", "properties": {...} }

create_or_replace_relationship

Create typed relationships between entities.

{ "source_id": "room-1", "target_id": "sensor-1", "relationship_name": "contains" }

update_digital_twin

Patch entity properties using JSON Patch operations.

{ "twin_id": "sensor-1", "patch": [{"op": "replace", "path": "/value", "value": 72}] }

Schema & Models

create_model

Define new DTDL models with properties and relationships.

{ "model": { "@id": "dtmi:example:Sensor;1", ... } }

get_model

Get full model definition including inherited properties.

{ "model_id": "dtmi:example:Sensor;1" }

list_models

List all models in the graph with summaries.

{}

Real-world patterns

See how agents use these tools to build and query knowledge graphs.

Semantic search for context

Use vector embeddings stored on entities to find semantically similar content. Filter by model type and limit results.

The search tool combines keyword matching with embedding similarity, giving you the best of both approaches.

semantic-search.jsJavaScript

// Semantic search for relevant context
await session.call_tool("search_digital_twins", {
    "search_text": "user authentication JWT validation",
    "model_id": "dtmi:code:Module;1",  // Optional filter
    "limit": 10
});

// Returns entities ranked by semantic similarity
// [
//   { "$dtId": "jwt-validator", "path": "src/auth/jwt.ts", ... },
//   { "$dtId": "session-handler", "path": "src/auth/session.ts", ... }
// ]

Traverse relationships

Use Cypher to navigate the graph structure. Find dependencies, trace lineage, or explore connections between entities.

Variable-length path matching lets you explore N-hop relationships without knowing the exact depth.

graph-traversal.jsJavaScript

// Traverse graph relationships with Cypher
await session.call_tool("query_digital_twins", {
    "query": `
        MATCH (root:Twin {name: 'api-gateway'})
        MATCH (root)-[:imports*1..3]->(dep:Twin)
        WHERE dep.\`$metadata\`.\`$model\` = 'dtmi:code:Module;1'
        RETURN DISTINCT dep.path, dep.description
        ORDER BY dep.path
    `
});

// Find all modules within 3 hops of api-gateway

Hybrid graph + vector queries

Combine structural graph queries with vector similarity ranking. Find entities that match a pattern AND are semantically similar.

This is the power of Konnektr: structural constraints from the graph with semantic ranking from embeddings.

hybrid-query.jsJavaScript

// Combine vector search with graph traversal
await session.call_tool("query_digital_twins", {
    "query": `
        MATCH (m:Twin)-[:imports]->(dep:Twin)
        WHERE dep.name = 'auth-service'
        RETURN m.path, m.description
        ORDER BY l2_distance(m.embedding, $query_vector)
        LIMIT 5
    `
});

// Find modules that import auth-service, ranked by similarity to query

Build custom agents

Use the Python MCP library to build agents that maintain their own knowledge graphs. Perfect for codebase understanding, document processing, or any domain where relationships matter.

pip install mcp

The official MCP Python library handles connection management and authentication automatically.

agent.pyPython

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async def build_codebase_knowledge(repo_path: str):
    url = "https://mcp.graph.konnektr.io/mcp?resource_id={resource_id}"

    async with streamablehttp_client(url) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()

            # Create a module entity in the graph
            result = await session.call_tool(
                "create_or_replace_digital_twin",
                arguments={
                    "twin_id": "module-auth-service",
                    "model_id": "dtmi:code:Module;1",
                    "properties": {
                        "path": "src/auth/service.ts",
                        "description": "Authentication service",
                        "embedding": embedding_vector
                    }
                }
            )

            # Query for related modules
            deps = await session.call_tool(
                "query_digital_twins",
                arguments={
                    "query": """
                        MATCH (m:Twin)-[:imports]->(auth:Twin)
                        WHERE auth.`$dtId` = 'module-auth-service'
                        RETURN m.path, m.description
                    """
                }
            )

Ready to connect your agents?

Deploy a graph and get MCP credentials in minutes.

Deploy Graph Read the Docs

Connect your agentsto structured memory