Claude Desktop + Claude Code

Memory for Claude

Persistent graph memory for Claude Desktop and Claude Code. MCP-native. Two-minute install.

The problem

Claude's context window is large, but it resets between sessions. Claude Projects help scope context to a project, but they do not let you cross-reference people, deals, or threads across projects. Graphory gives Claude a persistent graph to query and write to - a memory that survives restarts and grows with use.

Why Graphory with Claude

MCP-first

Claude was the launch partner for Model Context Protocol. Graphory's MCP integration is first-class.

Graph-based

A typed graph of the people, companies, activities, documents, accounts, and conversations in your business. Your Claude session can traverse real relationships.

Deterministic

No vectors, no embeddings, no hallucinated recall. Every query returns exact results.

Temporal

Every memory carries a timestamp, a source, and an audit trail. Claude can answer when and where.

Works in Claude Code too

The same MCP config works in Claude Code. Coding memory that persists across projects and restarts.

Two-minute install

Install the SDK, log in, register Graphory with Claude Desktop or Claude Code.

Step 1 - Install and authenticate

pip install graphory
graphory login

Step 2 - Add Graphory to claude_desktop_config.json

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "graphory": {
      "url": "https://api.graphory.io/mcp",
      "headers": {
        "Authorization": "Bearer gs_ak_your_key"
      }
    }
  }
}

Restart Claude Desktop. Graphory tools will appear in the MCP server list. Claude Code users: add the same block to your Claude Code MCP settings.

What you can do

Real queries Claude can run once Graphory is connected.

"Remember that Sarah from Acme mentioned pricing was their blocker in March."

"Show me everything I learned about Project X across all my Claude sessions this quarter."

"Who have I interacted with at companies in the fintech vertical?"

Measured, not marketed.

Deterministic extraction and retrieval, benchmarked against public datasets.

0.9107 accuracy

LongMemEval (agent + MCP)

0.8667 accuracy

LoCoMo-MC10 (agent + MCP)

0.900 F1

BizLineItemBench

Claude, with memory.

Free up to 100K nodes. No credit card.