v1.1.1-RC.1 Release

Einstein-MCP

Einstein MCP is a local MCP server with persistent knowledge management and experimental code-outline tools that map symbol relations across a codebase.

$bash -c "$(curl -fsSL https://cli.ben-fischer.dev/installer)" -- einstein-mcp
Einstein-MCP
$agent ~Extend the Role System to allow configuring permissions
Einstein: Recall General Memory
Einstein: Search Memories for authentication, roles, permission
Einstein: List Memories
Einstein: Recall Knowledge - UI structure
Great. I've got all information I need, let me start...
Einstein: Refine Memory - permission-management
Everything done. The new permission system is in place!

Overview

Overview

Einstein MCP is a local MCP server that stores durable project knowledge on disk so agents can recall, search, and refine memory across sessions. It also ships experimental code-outline tools that build a symbol relation graph from source code, letting agents navigate declarations, call chains, type hierarchies, and variable usages without leaving the MCP session.

  • File-based knowledge store rooted at a configurable project directory.
  • Tag-based search and topic-based retrieval with structured metadata.
  • Simple MCP toolset plus a setup prompt to bootstrap memory.
  • Experimental code-outline tools with language-aware symbol extraction (enabled via --experimental).

Available Tools

Einstein MCP provides 8 stable knowledge management tools and 3 experimental code-outline tools.

Knowledge — Read

  • recall_general_memory - Get the general project overview from the index
  • recall_memory - Retrieve detailed content for a specific topic
  • search_memories - Find topics by tags
  • list_memories - List all available memory topics

Knowledge — Write

  • memorize_new_topic - Create a new topic with tags and content
  • refine_memory - Update an existing topic's tags, summary, or content
  • memorize_general_info - Set or update the general project knowledge
  • forget_memory - Delete a topic and remove it from the index

Code Outline (experimental — requires --experimental flag)

  • outline_symbol - Relation tree for a named symbol across the project
  • outline_file - Relation tree for all symbols declared in a file
  • outline_project - Full relation tree for every symbol in the project

Setup Prompt

The memorize-project prompt provides step-by-step instructions for initializing a project's knowledge base:

  1. Analyze the project structure and configuration
  2. Create general project memory
  3. Identify and document specialized components
  4. Verify the memory index

Quick Start

Build the server:

go build -o bin/einstein-mcp ./cmd/mcp

Configure your MCP client (example for Claude Desktop):

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "einstein": {
      "command": "/absolute/path/to/einstein-mcp",
      "args": ["--root", "/path/to/your/project"]
    }
  }
}

Add "--experimental" to args to also enable the code-outline tools.

Claude Code MCP Configuration
$~claude mcp add einstein -- einstein-mcp --experimental
Adding MCP server "einstein" with command "einstein-mcp"...
MCP server "einstein" added successfully.

Initialize memory by invoking the memorize-project prompt in your MCP client, or manually call the tools to create your knowledge base.

USPs & Features

Features

Persistent knowledge store
Stores memories in a .ai-knowledge folder under the configured root.
Topic-based memory
Create, read, refine, and forget topics with YAML frontmatter.
Tag search
Find topics by tags using case-insensitive matching.
General project memory
Maintain a single general summary in index.mdk.
Prompted setup
Use the memorize-project prompt to initialize memory.
MCP-native
Built on mcp-go and served over stdio for easy client integration.
Code outline (experimental)
Build a symbol relation graph from source code — declarations, calls, type hierarchies, and variable usages across the whole project.
Language-aware extraction
Pluggable extractor architecture with a full Go AST extractor; additional languages can be added without changing the core.
Bidirectional relations
14 relation kinds including calledBy, accessedBy, usedBy, extendedBy, and implementedBy — queryable in both directions.
Documentation

Einstein-MCP Guides

Jump into the full documentation hub for setup, workflows, and references.

Open Docs

Latest Releases

View all releases
v1.1.1-RC.1
Mar 24, 2026
v1.1.1-RC.1
macOS Intel
Binary • macOS • amd64 • 8.0 MB
Download
macOS Apple Silicon
Binary • macOS • arm64 • 7.6 MB
Download
Linux AMD64
Binary • Linux • amd64 • 7.8 MB
Download
Linux ARM64
Binary • Linux • arm64 • 7.3 MB
Download
Windows AMD64
Binary • Windows • amd64 • .exe • 7.9 MB
Download

First release candidate for v1.1.1. Features may be unstable and buggy!

  • Added parsers for javascript, laravel, nextjs, php, react, symfony, typescript (mostly untested)
  • Added a parser for twig (roughly tested and adjusted)
  • Added automatic instructions describing how the memory management should be used
  • Added automatic instructions, describing how the outliner should be used

All features related to the outline feature are considered unstable and are only available via the --experimental flag!

v1.1.1-RC.2
Mar 24, 2026
v1.1.1-RC.2
macOS Intel
Binary • macOS • amd64 • 8.2 MB
Download
macOS Apple Silicon
Binary • macOS • arm64 • 7.8 MB
Download
Linux AMD64
Binary • Linux • amd64 • 8.0 MB
Download
Linux ARM64
Binary • Linux • arm64 • 7.5 MB
Download
Windows AMD64
Binary • Windows • amd64 • .exe • 8.1 MB
Download

Second release candidate for v1.1.1. May be unstable.

  • Added shopware-twig superset for the outline feature. This superset adds automatic plugin scanning and support for sw_include and sw_extends directives, to correctly work in Shopware projects.