Hypermode Graph MCP
Connect your Hypermode Graph to AI coding assistants using the Model Context Protocol for exploratory data analysis and intelligent graph operations
Overview
The Hypermode Graph MCP (Model Context Protocol) server enables integration
between your Hypermode Graph and AI coding assistants like Claude Desktop,
Cursor, and other MCP-compatible tools. Two MCP endpoints are available with
common tools for AI coding assistants: mcp
(an endpoint that provides tools
and data from your Graph) and mcp-ro
(an endpoint that provides tools and data
from your Graph in read-only mode).
What’s MCP? The Model Context Protocol is an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools. Think of the Hypermode Graph MCP server as a universal connector that allows AI assistants to understand and interact with your graph data.
Why use Hypermode Graph MCP?
For exploratory data analysis
- Interactive graph exploration: Query your knowledge graphs using natural language through your AI assistant
- Pattern Discovery: Let AI help identify relationships, clusters, and anomalies in your graph data
- Dynamic schema understanding: AI assistants can introspect your graph structure and suggest optimal queries
- Real-time insights: Get immediate answers about your data without writing complex DQL queries
For AI coding assistants
- Context-aware development: Your coding assistant understands your graph schema and data patterns
- Intelligent query generation: AI can write and optimize DQL queries based on your specific use case
- Schema evolution support: Get suggestions for schema changes and migrations
- Debugging assistance: AI can help troubleshoot graph queries and performance issues
Architecture
The Hypermode Graph MCP server follows the standard MCP architecture:
Components:
- MCP host: Your AI-powered app (Claude Desktop, IDE with AI features)
- MCP client: Protocol client that maintains connection with the Dgraph server
- Dgraph MCP server: Exposes graph capabilities through standardized MCP interface
- Graph database: Your Dgraph cluster (local or on Hypermode)
MCP server types
Hypermode Graph provides two MCP server variants to match your security and usage requirements:
Standard MCP server (mcp
)
- Full Access: Read and write operations on your graph
- Use Cases: Development environments, data modeling, schema evolution
- Capabilities: Query execution, mutations, schema modifications, namespace management
Read-only MCP server (mcp-ro
)
- Safe Exploration: Query-only access to your graph data
- Use Cases: Production analysis, reporting, exploratory data analysis
- Capabilities: DQL queries, schema introspection, data visualization support
Choose the read-only server (mcp-ro
) for production environments or when
working with sensitive data to prevent accidental modifications.
Setup guide
Running with Hypermode Graphs
When using Hypermode Graphs, the MCP configuration is available on the graph details screen in the console:
Access Graph Console
Navigate to your Hypermode workspace and select your graph instance.
Copy MCP Configuration
From the graph details screen, copy the provided MCP configuration.
Configure Your AI Assistant
Add the configuration to your AI assistant’s settings:
Running with local Dgraph
You can also run MCP with your local Dgraph instance by starting the Alpha
server with the --mcp
flag:
Starting Dgraph Alpha with MCP
This enables two MCP endpoints on your Alpha server:
- Full access:
http://localhost:8080/mcp/sse
- Read-only:
http://localhost:8080/mcp-ro/sse
Configure AI assistant for local Dgraph
Standalone MCP server
For development or testing, you can also run a standalone MCP server:
For standalone servers, configure your AI assistant with:
Available capabilities
Tools
Interactive tools for graph operations:
The Hypermode Graph MCP server provide the following tools:
get_schema
Retrieve the current schema of your graph database.
Example User request:
Tool call:
Response:
run_query
Run a DQL query on your graph database.
Parameters:
query
(string): DQL query to execute
Example User request:
Tool call:
Response:
run_mutation
Run a DQL mutation on your graph database.
Parameters:
mutation
(string): DQL mutation to execute
Example User request:
Tool call:
Response:
alter_schema
Modify the DQL schema of your graph database.
Parameters:
schema
(string): DQL schema to apply
Example User request:
Tool call:
Response:
get_common_queries
Provides reference queries to aid in query syntax.
Resources
The MCP server exposes read only data from your graph as resources:
Due to current limitations in some MCP clients that don’t yet support MCP resources, some tools are exposed as both tools and resources.
dgraph://schema
The current Dgraph DQL schema.
dgraph://common_queries
Pre-built query patterns for common operations.
Prompts
Pre-configured prompt templates for common graph operations:
quickstart_prompt
A quickstart prompt for getting started with graph MCP.
Example tool interactions
Here are additional examples of AI assistant interactions with the MCP tools using a financial transaction theme:
Complex transaction analysis
Example User request:
Tool call:
Response:
Account balance updates
Example User request:
Tool call:
Response:
Example workflows
Exploratory data analysis
Understand your data
Ask your AI assistant: “What does the graph schema look like and what are the main entity types?”
Discover patterns
“Find all users who have more than 10 connections and show their relationship patterns.”
Analyze distributions
“What’s the distribution of node types in the graph? Are there any outliers or interesting clusters?”
Performance insights
“Which queries are running slowly and how can we optimize them?”
AI-assisted development
Schema design
“I want to model a social network with users, posts, and interactions. What’s the optimal schema design?”
Query optimization
“This query is slow: [paste query]
. How can we improve its performance?”
Index recommendations
“Based on the query patterns, what indices should we add to improve performance?”
Migration planning
“I need to add a new ‘tags’ predicate to existing posts. What’s the safest migration approach?”
Best practices
Security
- Use Read-Only Servers: Default to
mcp-ro
for analysis and exploration - Authentication: Always use bearer tokens for Hypermode connections
Development workflow
- Start with Read-Only: Begin exploration with
mcp-ro
to understand your data - Iterative Schema Design: Use AI assistance for gradual schema evolution
- Query Testing: Validate AI-generated queries in development before production
- Documentation: Keep schema documentation updated for better AI understanding