memory-mcp
Personal memory MCP server for Claude Code with an Obsidian-style graph. Enables persistent, queryable memories across sessions using Markdown files with bidirectional wiki-links and a D3.js force-directed graph UI.
README
memory-mcp
Personal memory MCP server for Claude Code with an Obsidian-style interactive graph.
Persistent, queryable memory across Claude Code sessions. Memories are plain Markdown files with YAML frontmatter. Wiki-links ([[Memory Name]]) create bidirectional connections visualized in a D3.js force-directed graph at localhost:4242.
Screenshot
(graph UI screenshot here)
Features
- 8 MCP tools — save, read, search, list, delete, graph data, stats, open graph
- D3.js force-directed graph — Obsidian-style visual graph of all memories and their connections
- Wiki-links —
[[Memory Name]]in content auto-creates bidirectional links - Fuzzy search — powered by Fuse.js, searches name, content, and tags
- 5 memory types — each rendered as a distinct color in the graph
- Plain Markdown storage — memories are
.mdfiles with YAML frontmatter, readable without any tooling - Web UI at
localhost:4242— search, type filters, node click opens a sidebar with rendered Markdown - 96% test coverage — Vitest, isolated per module with temp vault directories
Quick Start
git clone https://github.com/otaviosenne/memory-mcp
cd memory-mcp
npm install
npm run build
Add to Claude Code's ~/.claude/mcp.json:
{
"mcpServers": {
"memory-mcp": {
"type": "stdio",
"command": "node",
"args": ["/path/to/memory-mcp/dist/index.js"],
"env": {
"MEMORY_VAULT_PATH": "~/.local/share/memory-mcp/vault",
"MEMORY_WEB_PORT": "4242"
}
}
}
}
Restart Claude Code. The MCP server starts automatically with the session.
MCP Tools
| Tool | Description |
|---|---|
memory_save |
Create or update a memory by name |
memory_read |
Read a single memory by ID or name |
memory_search |
Fuzzy search across name, content, and tags |
memory_list |
List all memories, optionally filtered by type |
memory_delete |
Delete a memory by ID |
memory_graph_data |
Return all nodes and edges for graph rendering |
memory_stats |
Count memories by type, total links, vault size |
memory_open_graph |
Open the web UI in the default browser |
Example: memory_save
Input:
{
"name": "Chrome Agent MCP",
"type": "project",
"content": "Browser automation MCP using Chrome DevTools Protocol. Built on [[memory-mcp]] for session context.",
"tags": ["mcp", "typescript", "chrome"]
}
Output:
{
"id": "a1b2c3d4-...",
"name": "Chrome Agent MCP",
"type": "project",
"links": ["<uuid-of-memory-mcp>"],
"created": "2026-03-24T10:00:00.000Z"
}
Example: memory_search
Input:
{ "query": "typescript mcp" }
Output: array of matching memories, sorted by relevance score.
Example: memory_stats
Output:
{
"total": 24,
"byType": { "project": 10, "user": 5, "feedback": 4, "reference": 3, "note": 2 },
"totalLinks": 38
}
Memory Types
| Type | Graph Color | When to Use |
|---|---|---|
user |
Green | Personal info, preferences, skills, background |
project |
Blue | Project decisions, tech stack, current status, constraints |
feedback |
Yellow | Corrections and confirmed patterns ("don't do X", "yes, exactly like this") |
reference |
Pink | External resources, accounts, tools, links |
note |
Gray | General context that doesn't fit other types |
Wiki-links
Use [[Memory Name]] syntax anywhere in a memory's content to reference another memory by exact name.
This project depends on [[memory-mcp]] for persistent context across sessions.
The architecture follows the patterns described in [[SOLID Design Principles]].
Links are resolved at save time. Both the source and target memory store the connection — the graph is always bidirectional. If the target memory does not exist yet, the link is stored as unresolved and re-evaluated when the target is created.
Graph UI
Open localhost:4242 while the MCP server is running (or call memory_open_graph).
- Node size scales with connection count — highly-linked memories appear larger
- Node color maps to memory type
- Click a node to open a sidebar with the full Markdown content, rendered
- Search bar filters nodes by name or tag in real time
- Type toggles show or hide entire memory types
- Zoom and pan are supported via D3 zoom behavior
The graph updates live — refresh the page to reflect new memories.
Claude Code Skill: /save-memories
At the end of a session, run /save-memories to extract and persist meaningful context automatically.
The skill reviews the conversation and decides what is worth saving. It skips trivial fixes and ephemeral tasks — it only saves things that would be genuinely useful in a future session.
~/.claude/skills/save-memories/SKILL.md:
---
name: save-memories
description: Analyze the current conversation and save meaningful memories to the personal memory vault. Use at the end of any session where something worth remembering was discussed — new projects, preferences revealed, corrections made, important context.
user-invocable: true
---
# Save Memories
Analyze the current conversation and save relevant memories to the vault at `/home/senne/.local/share/memory-mcp/vault/`.
## Step 1 — Review the conversation
Read back through what was discussed. Ask: what here would be useful to know in a future conversation that isn't already obvious from the code?
## Step 2 — Decide what to save
Only save if it reveals:
| Type | Save when |
|------|-----------|
| `user` | Personal info, skills, preferences, opinions, background |
| `feedback` | Corrections ("not like that"), confirmed patterns ("yes, exactly") |
| `project` | Project decisions, stack, current status, key constraints |
| `reference` | External resources, accounts, tools, links, credentials context |
| `note` | Important context that doesn't fit other types |
**Skip:** trivial bug fixes, ephemeral tasks, things obvious from the code, duplicates.
## Step 3 — Save each memory
For each memory, run:
```bash
node -e "
import('/path/to/memory-mcp/dist/core/vault.js').then(async ({ Vault }) => {
const vault = new Vault('/path/to/vault');
await vault.initialize();
const saved = await vault.save({
name: 'MEMORY NAME',
type: 'project',
content: 'Content here. Use [[Other Memory Name]] to link related memories.',
tags: ['tag1', 'tag2'],
});
console.log('Saved:', saved.name, '| links:', saved.links.length);
process.exit(0);
});
" 2>/dev/null
Use [[wiki-links]] in content to connect related memories. The exact name must match an existing memory name.
Step 4 — Report
After saving, briefly list what was saved and why. Keep it to one line per memory.
---
## Architecture
src/ ├── index.ts # MCP server entry point, tool registration ├── types/index.ts # All TypeScript types and interfaces ├── core/ │ ├── storage.ts # File I/O abstraction (read, write, delete, list) │ ├── parser.ts # Frontmatter parsing and serialization (gray-matter) │ ├── linker.ts # Wiki-link extraction and bidirectional resolution │ └── vault.ts # Orchestrator: CRUD, search, graph, stats ├── tools/ # 8 MCP tool handlers, one file per tool └── web/ ├── server.ts # Express REST API serving graph data └── public/index.html # D3.js graph UI, single-file, no build step
Each module has one responsibility. The `Vault` class composes `Storage`, `Parser`, and `Linker` — it does not implement any of their logic directly.
---
## Tech Stack
| Package | Role |
|---------|------|
| `@modelcontextprotocol/sdk` | MCP server protocol |
| `d3` v7 | Force-directed graph in the browser |
| `express` | REST API for the web UI |
| `gray-matter` | YAML frontmatter parsing |
| `fuse.js` | Fuzzy search |
| `zod` | Tool input validation |
| `vitest` | Tests (96% coverage) |
| TypeScript strict mode | Type safety throughout |
---
## Code Patterns & SOLID
The codebase applies SOLID principles throughout:
**Single Responsibility** — each module does exactly one thing. `Storage` handles file I/O. `Parser` handles Markdown/frontmatter. `Linker` handles wiki-link extraction and resolution. `Vault` orchestrates them.
**Open/Closed** — new memory types can be added by extending the `MemoryType` union in `types/index.ts`. No core logic changes.
**Dependency Inversion** — `Vault` depends on the `Storage` and `Parser` interfaces, not their concrete implementations. This makes the core fully testable with mock implementations.
**Small files** — every file stays under 100 lines. No file mixes concerns.
No comments in the source — names are the documentation. No magic numbers — all constants are named. All code is in English.
---
## Testing
```bash
npm test # run all tests
npm run test:coverage # coverage report (target: 70%+, actual: 96%)
Tests live in tests/, mirroring src/:
tests/
├── core/
│ ├── storage.test.ts
│ ├── parser.test.ts
│ ├── linker.test.ts
│ └── vault.test.ts
└── tools/
└── *.test.ts
Each test suite creates a temporary vault directory and tears it down after. No shared state between tests.
Environment Variables
| Variable | Default | Description |
|---|---|---|
MEMORY_VAULT_PATH |
~/.local/share/memory-mcp/vault |
Directory where memory .md files are stored |
MEMORY_WEB_PORT |
4242 |
Port for the graph web UI |
Memory File Format
Each memory is stored as a .md file with YAML frontmatter:
---
id: a1b2c3d4-e5f6-7890-abcd-ef1234567890
name: Chrome Agent MCP
type: project
tags:
- mcp
- typescript
- chrome
created: 2026-03-24T10:00:00.000Z
updated: 2026-03-24T10:00:00.000Z
links:
- b2c3d4e5-f6a7-8901-bcde-f12345678901
---
Browser automation MCP using Chrome DevTools Protocol.
Built on [[memory-mcp]] for persistent session context.
Supports screenshot, click, fill, scroll, and JS evaluation.
Files are human-readable and can be edited directly. The links array is managed automatically by the linker — manual edits to links are valid but wiki-links in content take precedence on next save.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.