MemoryGraph - Memory for AI Coding Agents | Never Re-explain Your Project Again

> SEE IT IN ACTION_

terminal

SESSION 1:

You: "Remember: retry with exponential backoff fixed the API timeout issues"

Claude: ✓ Stored in memory

─────────────────────────────────────────────

SESSION 2 (days later):

You: "What fixed the timeout issues we had?"

Claude: "We fixed the API timeouts using exponential backoff retry logic. This solved both the connection timeouts and rate limiting issues."

> THE PROBLEM_

✗ WITHOUT MEMORY

Session 1: "Let me explain my project architecture..."

Session 2: "Let me explain my project architecture..." (again)

Session 3: "Let me explain my project architecture..." (seriously?)

Session 47: *quiet sobbing*

✓ WITH MEMORYGRAPH

Session 1: "Remember this architecture..." ✓ Stored

Session 2: Claude already knows. You ship.

Session 3: Claude recalls what worked AND what failed. You ship faster.

Session 47: Claude knows your project better than you remember it yourself.

> FEATURES_

PERSISTENT MEMORY

□ ×

Store solutions, patterns, decisions, and learnings that survive across sessions. Never lose context to compaction again.

> "Remember: the auth bug was caused by token expiry" > "Remember: we use repository pattern for data access" > "Remember: pytest fixtures go in conftest.py"

KNOWS WHAT SOLVED WHAT

□ ×

Not just notes—connected knowledge. MemoryGraph tracks relationships between problems, solutions, and patterns.

RetryBackoff

──SOLVES──▶

APITimeout

Ask "what solved the timeout?" and get the actual solution, not just text that mentions "timeout" somewhere.

ZERO CONFIG

□ ×

Works in 30 seconds. SQLite by default—no database to set up, no cloud account to create, no API keys to manage.

$ pipx install memorygraphMCP $ claude mcp add memorygraph -- memorygraph # That's it. You're done.

Your data stays on your machine. Works offline. Privacy by default.

PROJECT AWARE

□ ×

Organize memories by project. Switch contexts cleanly. What you learned on Project A doesn't pollute Project B.

> Project: api-service └── 47 memories (auth, caching, deployment)

> Project: data-pipeline └── 23 memories (ETL patterns, error handling)

> Project: personal-site └── 12 memories (styling, deployment)

> HOW IT WORKS_

01 CONFIGURE

Add this to your CLAUDE.md for automatic memory storage:

## Memory Protocol

REQUIRED: Before Starting Work

You MUST use `recall_memories` before any task.

REQUIRED: Automatic Storage Triggers

• Git commit → what was fixed/added

• Bug fix → problem + solution

• Version release → summarize changes

• Pattern discovered → reusable approach

Do NOT wait to be asked. Storage is automatic.

02 STORE

Just talk naturally. Tell Claude what to remember.

> "Remember: exponential backoff fixed the API timeouts. Start at 100ms, max 5 retries, 2x multiplier."

↓

Entity: ExponentialBackoff

Type: Solution

Observations:

- Start at 100ms delay

- Maximum 5 retries

- 2x multiplier between attempts

Relationship: SOLVES → APITimeout

03 RECALL

Ask about past solutions. Claude searches your memory.

> "What fixed the timeout issues we had?"

↓

Claude searches memory → Finds ExponentialBackoff → Sees it SOLVES APITimeout → Returns the full context

Claude: "We fixed the API timeouts using exponential backoff. The configuration was: 100ms initial delay, max 5 retries, 2x multiplier between attempts."

04 BUILD ON IT

Knowledge compounds. Claude gets smarter about YOUR project.

Week 1: Claude learns your architecture

Week 2: Claude learns your patterns

Week 3: Claude learns what works and what doesn't

Week 4: Claude suggests solutions based on what worked before

Your AI assistant evolves from "helpful stranger" to "team member who's been on the project for months."

> VS THE ALTERNATIVES_

VS CLAUDE.md

CLAUDE.md is great for static instructions:
"Use pytest for testing. Follow PEP 8."

MemoryGraph is for dynamic learnings:
"The caching bug was fixed by clearing Redis on deploy."
"The auth tokens expire after 1 hour, not 24."
"Don't use that library—it broke production last month."

CLAUDE.md = what to do (instructions)

MemoryGraph = what we learned (knowledge)

Use both. They complement each other.

VS BASIC-MEMORY

basic-memory is excellent general-purpose PKM.
Markdown files, Obsidian integration, cloud sync.

MemoryGraph is purpose-built for coding workflows:

Typed relationships (SOLVES, CAUSES, BLOCKED_BY)
"What solved this?" not just "what mentions this?"
Designed for how developers actually work

basic-memory = "remember what I said"

MemoryGraph = "remember what actually worked"

VS ANTHROPIC'S BUILT-IN MEMORY

Anthropic's memory is convenient—zero setup.

MemoryGraph gives you:

✓ Control: See exactly what's stored
✓ Portability: Export, backup, move your data
✓ Privacy: Data never leaves your machine
✓ Structure: Relationships, not just facts
✓ Project separation: Clean context switching

Your knowledge. Your machine. Your control.

VS DOING NOTHING

The cost of no memory:

Re-explaining your architecture every session
Re-discovering solutions you already found
Re-making mistakes you already made
Context lost every time the window fills up

Time spent re-explaining = time not shipping.

MemoryGraph installs in 30 seconds.
How much time will you save this week?

> FREQUENTLY ASKED QUESTIONS_

> How is my data stored?

SQLite database on your local machine by default. Your data never leaves your computer unless you explicitly configure cloud sync (coming soon).

> Does this work with Claude Desktop or just Claude Code?

Works with any MCP-compatible client: Claude Code, Claude Desktop, Cursor, Windsurf, VS Code + Copilot, Continue.dev, Cline, and more.

> Can I use this with multiple projects?

Yes. MemoryGraph supports project-based organization. Each project has its own memory space.

> What happens if I want to switch to a different tool later?

Your data is yours. Export to JSON or Markdown anytime. No lock-in.

> Is there a cloud/team version?

Coming soon. The free local version will always exist. Cloud sync and team features will be paid tiers. Star the repo for updates.

> How does this compare to RAG?

RAG retrieves from static documents. MemoryGraph captures dynamic learnings from your conversations and tracks relationships between them. Different tools for different jobs.

> QUICK START_

bash

# 1. Install

$ pipx install memorygraphMCP

# 2. Add to Claude Code

$ claude mcp add --scope user memorygraph -- memorygraph

# 3. Restart Claude Code (exit and run 'claude' again)

# 4. Verify

$ claude mcp list

memorygraph: Connected

~/.cursor/mcp.json

# 1. Install

$ pipx install memorygraphMCP

# 2. Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "memorygraph": {
      "command": "memorygraph"
    }
  }
}

# 3. Restart Cursor

mcp.json

# Generic MCP configuration

{
  "mcpServers": {
    "memorygraph": {
      "command": "memorygraph",
      "args": ["--profile", "extended"]
    }
  }
}

# See docs for: Windsurf, VS Code, Continue, Cline, Gemini CLI