M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
M

mcp-knowledge-graph

Community
Persistent memory for AI models through a local knowledge graph.
How to install ↓ Compare GitHub source repository ↗
Category
Database #28 of 58
Stars
★ 888 Very popular
Transport
stdio (local process)
Runtime
Node.js 22+
Credentials
No credential needed
License
MIT
Last commit
Tools
10
56FMRS · C

This server offers a flexible and safe design for AI memory management, supporting project-local and global storage with multi-database features. Installation is straightforward, though compatibility with Claude API needs verification.

Strongest · Documentation 16/20 Weakest · Reliability 7/20

Reliability
7/20
Security and permissions
11/20
Maintenance
8/20
Documentation
16/20
Setup experience
14/20
Why each score
Reliability 7/20
Evidence: README describes ten tools, a storage-priority model, the '_aim' file-marker flow, and troubleshooting with concrete error text; the design is internally coherent. Deductions: the server manifest is not cached and no source files, committed tests, or CI workflows were supplied, so the MCP init/tool-list handshake and real tool behavior cannot be verified. Static calibration caps reliability at 12 without execution evidence, and uncertainty lowers it further.
Security and permissions 11/20
No red-line issues: no credential handling, no network exfiltration, no real tokens in install examples, no destructive defaults. README discloses data boundaries — writes go only to a project '.aim' directory or the configured '--memory-path' directory — and describes an '_aim' marker that refuses to overwrite unrelated JSONL files. Deductions: marker enforcement cannot be verified in source, 'forget'/'remove' operations have no described confirmation beyond the client's own approval, and project auto-detection can surprise users about where writes land. Least-privilege design is plausible but unproven.
Maintenance 8/20
Facts available: MIT license, repository not archived, only 1 open issue, described as a fork focused on local development. Deductions: the supplied material contains no commit history, release tags, dependency-update records, issue-response evidence, or security-response channel; the server manifest is not cached. Stars are a discovery signal only and were not scored. Sustained maintenance cannot be confirmed.
Documentation 16/20
README is layered: quick start for global and project-local modes, two copy-paste client configs, a full tool list with 'context'/'location' parameters, an explicit explanation of '.aim' vs '_aim', file-organization trees, database discovery, autoapprove guidance, troubleshooting, requirements, and license. Limitations and failure modes are disclosed. Deductions: no dedicated auth/security notes (auth is N/A for a local file server but not stated), no explicit scale or cost limits, and documented behavior lacks source-backed verification.
Setup experience 14/20
Clear few-step setup: run 'npx -y mcp-knowledge-graph' with optional '--memory-path', paste one of the two ready-made JSON configs into Claude Desktop/.claude.json, or create an '.aim' directory for project-local mode; Node.js 22+ is stated. Deductions: static calibration caps setup at 15 when no CI workflows or committed tests verify the install path, and the npx package publication plus runtime behavior cannot be confirmed from the provided material.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local files

Best for

  • Claude users who need persistent memory across conversations
  • Developers who want local persistent memory in their projects
  • Users concerned about data privacy and prefer local storage over cloud

Not for

  • Scenarios requiring distributed or high-concurrency storage
  • Advanced security needs like built-in encryption or access control
  • Users who wish to store memory files in the cloud managed by a service provider

Required permissions

  • Read and write to the local file system (within specified --memory-path or project .aim directory)
  • Run npx commands to start the server

Risks and side effects

  • Local files might be accidentally deleted or overwritten (the system has _aim marker protection, but risk remains)
  • Sensitive information stored locally could be exposed if not encrypted
  • Dependency on Node.js version and npm package management may cause conflicts

Setup

Before you start

Runtime：Node.js 22+

Prerequisites: Node.js 22+. Setup: Add the MCP server configuration to your Claude client configuration file, specifying a local path (e.g., --memory-path), then start the client to automatically install and run.

claude_desktop_config.json 
{
  "mcpServers": {
    "Aim-Memory-Bank": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-knowledge-graph",
        "--memory-path",
        "/Users/yourusername/.aim"
      ],
      "autoapprove": [
        "aim_memory_search",
        "aim_memory_get",
        "aim_memory_read_all",
        "aim_memory_list_stores"
      ]
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "Aim-Memory-Bank": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-knowledge-graph",
        "--memory-path",
        "/Users/yourusername/.aim"
      ],
      "autoapprove": [
        "aim_memory_search",
        "aim_memory_get",
        "aim_memory_read_all",
        "aim_memory_list_stores"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add Aim-Memory-Bank -- npx -y mcp-knowledge-graph --memory-path /Users/yourusername/.aim

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Tools prefixed with aim_ such as aim_memory_store and aim_memory_search should appear in your client's tool list; calling aim_memory_list_stores and getting back the default master database confirms the connection works.

Troubleshooting

  1. If you encounter 'File does not contain required _aim safety marker' error, ensure the file starts with the correct marker or delete the file and let the system recreate it.
  2. If memories appear in unexpected locations, check if you are in a project directory with a .aim folder, and use aim_memory_list_stores to see the current storage location.
  3. If too many similar databases, manually delete unwanted .jsonl files and encourage the AI to use consistent naming.

Things to try

Once connected, you can ask your AI assistant things like:

  • Remember that John_Doe is a person I met at a conference
  • Search the work database for memories about the Q4 project
  • List all available memory databases
  • Forget the memory about the old project in my personal database

Tools 10

aim_memory_store writes
Stores new memories (people, projects, concepts) into the knowledge graph.
aim_memory_add_facts writes
Adds facts to existing memories.
aim_memory_link writes
Links two memories together.
aim_memory_search read-only
Searches memories by keyword.
aim_memory_get read-only
Retrieves specific memories by exact name.
aim_memory_read_all read-only
Reads all memories in a database.
aim_memory_list_stores read-only
Lists all available databases.
aim_memory_forget destructive
Forgets memories.
Show 2 more tools
aim_memory_remove_facts destructive
Removes specific facts from a memory.
aim_memory_unlink destructive
Removes links between memories.

Use cases

Personal memory management: Let AI assistants remember your preferences, birthdays, project info, etc.
Project-local memory: Create a .aim folder in your project directory to automatically use project-level memory.
Cross-device sync: Use a cloud-synced directory like Dropbox to store memory files for multi-device access.

Supported clients

Claude Desktop
Claude Code

Listed from the project's documentation, not tested by this site.

Overview

mcp-knowledge-graph is an MCP server that enables AI models (like Claude) to store and retrieve information across conversations. It uses entities, relations, and observations to build a knowledge graph, and supports multiple databases (master and named) as well as project-local or global storage. The server emphasizes safety with file markers (_aim) to prevent accidental overwrites of unrelated files. It is designed for local development, suitable for users who want their AI assistant to remember long-term information.

Similar servers

Compare with other MCP servers

Source revision 9f6268c4b5df Data synced 2026-10-12 Read the FMRS scoring method