Connect AI tools with MCP :: TigerGraph Savanna

TigerGraph MCP connects AI development tools to your TigerGraph database, allowing agents to manage graphs, explore schemas, query graph data, load data, and perform database operations using MCP tools.

This page explains what MCP is, what TigerGraph MCP server is, and gets you from install to a successful tool call in Cursor, VS Code, Claude Code, or Claude Desktop.

What is MCP?

The Model Context Protocol (MCP) is an open standard that connects AI applications to external systems through a common interface. Instead of hardcoding a custom integration for every service, an MCP client (your AI tool) connects to an MCP server that exposes capabilities as tools the model can call.

When you ask the agent to do something, it chooses which tools to call, runs them, and uses the results to complete your request. Because MCP is standardized, the same server works across any MCP-capable client, such as Cursor, VS Code, Claude Code, and Claude Desktop.

How it works in practice:

The AI client decides when to call a tool based on your prompt and the conversation, so you do not have to invoke tools manually.
Each tool call runs during response generation, so the agent works with live data from the connected system instead of relying on training data or guesswork.
Tools can both read and act, so an agent can inspect a system and, when you ask, change it.

What is TigerGraph MCP server?

The TigerGraph MCP server is TigerGraph’s implementation of the Model Context Protocol, distributed as the tigergraph-mcp Python package on PyPI. It is the server your AI tool (the MCP client) connects to. It runs locally as a command-based server that your MCP client launches, and turns TigerGraph database operations into MCP tools your agent can call: creating and managing graphs, exploring schemas, reading and writing vertices and edges, running GSQL, creating loading jobs, and working with vector search.

Once connected, your AI tool works with your Savanna database in natural language. You describe what you want; the agent picks the right TigerGraph tools, runs them against the database you configured, and returns the results, with no hand-written REST calls or pasted JSON.

For how the pieces fit together, see How it works.

What you can do

TigerGraph MCP gives your agent hands-on access to the database, so you work in plain language instead of REST calls and GSQL boilerplate. From a single prompt, the agent can:

Design and evolve graphs. Create graphs, read and change schemas, and reshape the data model as your requirements shift.
Read and write graph data. Fetch vertices and edges, traverse neighbors, and upsert new data without leaving the chat.
Query the graph. Ask a question in plain English; the agent writes and runs the GSQL, fetches the results, and explains them back to you. It can also install and reuse queries for repeatable, production-grade workflows.
Load at scale. Create loading jobs, pull from the sources Savanna supports, and track job status as data lands.
Power vector search. Manage vector attributes, upsert embeddings, and run top-k similarity search for RAG and recommendations.
Check its own work. Inspect vertex, edge, and degree counts, and discover the tools and workflows it needs for multi-step tasks.

Under the hood that is dozens of tools, but you never call them directly. You describe the goal; the agent picks the tools, runs them against your database, and reports back with real results.

Before you start

A running Savanna workspace with an attached database. If you do not have one yet, build a graph in the console first.
A database secret for that database. See Create a database secret.
uv installed, if you use the recommended uvx path.

How the server runs

You do not start the server yourself. Your MCP client launches it from the command in your configuration.

The TigerGraph MCP server is distributed as a Python package. The recommended command uses uvx (bundled with uv), which runs it in an isolated environment with no separate install, the way npx runs npm tools.

"command": "uvx",
"args": ["tigergraph-mcp"]

Prefer to install it yourself? Run pip install tigergraph-mcp and set "command": "tigergraph-mcp" instead.

Configure your connection

Set three environment variables in your MCP client configuration.

Variable Required What to use

Variable	Required	What to use
`TG_HOST`	Yes	Your Savanna workspace URL. Open Workspaces, select the workspace, and copy its URL.
`TG_SECRET`	Yes	A database secret for that workspace’s database. See Create a database secret.
`TG_GRAPHNAME`	No	The graph to use by default. Set it if most of your work is on one graph. Leave it out if you work across several. You can still pass a different graph on any single tool call.

TG_HOST

Yes

Your Savanna workspace URL. Open Workspaces, select the workspace, and copy its URL.

TG_SECRET

Yes

A database secret for that workspace’s database. See Create a database secret.

TG_GRAPHNAME

The graph to use by default. Set it if most of your work is on one graph. Leave it out if you work across several. You can still pass a different graph on any single tool call.

Connect your AI tool

Pick your client, install the server, then fill in your own values. The install links carry placeholders only, so you never enter credentials on this page.

Cursor
Claude Desktop
VS Code
Claude Code

To open Cursor and add the TigerGraph MCP server automatically, select Add to Cursor. You can also add the following to your ~/.cursor/mcp.json file. To learn more, see the Cursor documentation.

{
  "mcpServers": {
    "tigergraph": {
      "command": "uvx",
      "args": ["tigergraph-mcp"],
      "env": {
        "TG_HOST": "YOUR_WORKSPACE_URL",
        "TG_GRAPHNAME": "YOUR_GRAPH_NAME",
        "TG_SECRET": "YOUR_DATABASE_SECRET"
      }
    }
  }
}

Add the following to your Claude Desktop config file. On macOS open ~/Library/Application Support/Claude/claude_desktop_config.json, and on Windows open %APPDATA%\Claude\claude_desktop_config.json. You can also use Settings → Developer → Edit Config. Save the file, then restart Claude Desktop fully. To learn more, see the Claude Desktop documentation.

{
  "mcpServers": {
    "tigergraph": {
      "command": "uvx",
      "args": ["tigergraph-mcp"],
      "env": {
        "TG_HOST": "YOUR_WORKSPACE_URL",
        "TG_GRAPHNAME": "YOUR_GRAPH_NAME",
        "TG_SECRET": "YOUR_DATABASE_SECRET"
      }
    }
  }
}

To open VS Code and add the TigerGraph MCP server automatically, select Install Server. You can also add the following to a workspace file at .vscode/mcp.json, or run MCP: Open User Configuration for a user level file. To learn more, see the VS Code documentation.

{
  "servers": {
    "tigergraph": {
      "type": "stdio",
      "command": "uvx",
      "args": ["tigergraph-mcp"],
      "env": {
        "TG_HOST": "YOUR_WORKSPACE_URL",
        "TG_GRAPHNAME": "YOUR_GRAPH_NAME",
        "TG_SECRET": "YOUR_DATABASE_SECRET"
      }
    }
  }
}

Add the TigerGraph MCP server from the terminal, then restart Claude Code.

claude mcp add --transport stdio tigergraph \
  --env TG_HOST=YOUR_WORKSPACE_URL \
  --env TG_GRAPHNAME=YOUR_GRAPH_NAME \
  --env TG_SECRET=YOUR_DATABASE_SECRET \
  -- uvx tigergraph-mcp

You can also add the following to a project file at .mcp.json. To learn more, see the Claude Code documentation.

{
  "mcpServers": {
    "tigergraph": {
      "command": "uvx",
      "args": ["tigergraph-mcp"],
      "env": {
        "TG_HOST": "YOUR_WORKSPACE_URL",
        "TG_GRAPHNAME": "YOUR_GRAPH_NAME",
        "TG_SECRET": "YOUR_DATABASE_SECRET"
      }
    }
  }
}

Once the TigerGraph MCP server is configured with the values from Configure your connection, ask your agent:

List the graphs available in my TigerGraph database.

If the agent returns graph information, the connection is working.

Try TigerGraph MCP

After the connection works, try these prompts. Start with read-only requests.

Explore your database

List the graphs available in my TigerGraph database.

Show the schema of my graph and explain how the vertex types are connected.

Show the vertex and edge counts for my graph.

Explore graph data

Find the neighbors of a sample vertex and explain the relationships.

Show me a few sample vertices from this graph.

Query the graph

Write a GSQL query to find the top 10 vertices by degree.

Review the query before allowing the agent to install or run it.

Load data

Show me the TigerGraph MCP tools available for loading data.

Help me create a loading job for this CSV file.

Discover MCP capabilities

What TigerGraph tools do you have available?

How it works

The TigerGraph MCP server runs locally as a command-based MCP server:

Cursor / VS Code / Claude Code / Claude Desktop
              |
              | MCP over stdio
              v
       tigergraph-mcp
              |
              | pyTigerGraph
              v
      TigerGraph Savanna

The AI application launches tigergraph-mcp on your machine.
The TigerGraph MCP server exposes TigerGraph operations as MCP tools.
The AI application decides which tools to call based on your request.
The TigerGraph MCP server uses your configured Savanna connection details to perform the operations.
Tool results return to the AI application and appear in its response.

Tool behavior

Your AI client determines which TigerGraph MCP tools to use based on your prompt and the current conversation.

Example:

"Show me my graph schema"
          ↓
AI chooses a schema tool
          ↓
TigerGraph MCP server
          ↓
Savanna

For a more complex task such as "Load this customer CSV and verify the data", the agent may:

Inspect the graph schema
Discover loading tools
Create or run a loading job
Check vertex and edge counts
Return the result

Actions performed through TigerGraph MCP run against the connected database and can modify or permanently delete data and graph resources. Review destructive tool calls carefully before approving them, particularly in production.

Troubleshooting

`uvx` is not found

macOS:

which uvx

Windows:

where.exe uvx

If the MCP client cannot resolve uvx, set command to the absolute path returned by your shell.

`tigergraph-mcp` is not found

Verify the package is installed:

pip install tigergraph-mcp

Then locate the executable:

macOS:

which tigergraph-mcp

Windows:

where.exe tigergraph-mcp

Use that absolute path as command if the client cannot find it on PATH.

Works in the terminal but not in the MCP client

Desktop applications can use a different PATH than your terminal. Use the full executable path in command.

Examples:

"command": "/absolute/path/to/uvx"

"command": "C:\\absolute\\path\\to\\uvx.exe"

Do not assume a fixed install directory. The same approach applies to tigergraph-mcp.

MCP starts but cannot connect

Check that:

TG_HOST points to the intended Savanna workspace
TG_SECRET is valid for the intended database
TG_GRAPHNAME, if present, matches a graph in that database
You remove TG_GRAPHNAME when a graph-specific default is unnecessary
The selected workspace and database are available

Windows JSON paths

Escape backslashes in JSON:

"command": "C:\\path\\to\\uvx.exe"

Feedback and contributions

Found a bug or unexpected behavior? Open an issue in the TigerGraph MCP GitHub repository. If you have a fix, submit a pull request.

Same outcomes without an agent: Connect via APIs
Console snippet generator: Connect via APIs
Load data in the console: Load data
Write queries in the console: GSQL Editor

Connect AI tools with MCP