Introduction to MCB

Relevant source files

The following files were used as context for generating this wiki page:

• README.md
• AGENTS.md
• crates/mcb-domain/src/lib.rs
• crates/mcb-server/Cargo.toml
• crates/mcb-infrastructure/Cargo.toml
• config/mcb-validate.toml
• crates/mcb-server/src/controllers/web/mod.rs

Memory Context Browser (MCB) is a high-performance Model Context Protocol (MCP) server implemented in Rust. It provides AI coding agents with persistent memory, semantic code search, and codebase understanding capabilities. By utilizing vector embeddings and tree-sitter parsing, MCB enables agents to maintain cross-session observations and perform deep analysis of multi-language repositories.

The system is built on a "Clean Architecture" foundation, ensuring strict separation between domain logic, infrastructure, and transport layers. It supports various embedding providers (e.g., OpenAI, Ollama) and vector stores (e.g., Qdrant, Milvus), making it adaptable to both local and cloud-based AI workflows.

Sources: README.md:1-21, AGENTS.md:154-156

Architecture Overview

MCB follows a strict Clean Architecture pattern where dependencies flow inward toward the domain core. This structure is enforced at compile time through workspace crates and automated validation rules.

Core Crate Layers

The project is divided into several primary crates, each representing a specific architectural layer:

Crate	Responsibility	Dependency Direction
mcb	CLI facade and entry point binary	Depends on all layers
mcb-server	MCP protocol, transport, and web controllers	Inward to Infrastructure/Domain
mcb-infrastructure	DI, configuration, cache, and logging	Inward to Domain
mcb-domain	Core entities, value objects, and port traits	Zero internal dependencies
mcb-providers	Adapters for specific technologies (Git, Vector stores)	Implements Domain ports
mcb-utils	Shared leaf utilities and constants	Shared by all

Sources: README.md:73-88, AGENTS.md:213-228, crates/mcb-domain/src/lib.rs:13-28

Dependency Flow Diagram

The following diagram illustrates the inward-flowing dependency model used by MCB to maintain architectural integrity.

MCP Tools and Capabilities

MCB exposes its functionality through the MCP protocol, allowing clients like Claude Desktop to invoke specific tools. These tools are grouped into nine operation families.

Key Tool Families

• Search: Includes search_code and search_memory for natural language queries.
• Index: Operations like index_repo and clear_index for managing vector embeddings of codebases.
• Memory: Facilities for store_memory and inject_context to handle persistent cross-session observations.
• Validate: Architecture and code analysis via validate_code.
• VCS: Git-aware tools for compare_branches and analyze_impact.

Sources: README.md:55-69, AGENTS.md:255-272

Tool Call Sequence

This diagram shows how a tool request flows from an MCP client through the server to the domain logic.

Infrastructure and Configuration

MCB uses a multi-provider strategy for embeddings and storage, configurable via environment variables or YAML files.

Provider Support

MCB supports a wide range of integrations:

• Embedding Providers: OpenAI, VoyageAI, Ollama, Gemini, FastEmbed, Anthropic.
• Vector Stores: Milvus, EdgeVec, Qdrant, Pinecone, Encrypted.
• AST Parsing: Supported for 14 languages including Rust, Python, Go, and Java.

Sources: README.md:9-15, crates/mcb-infrastructure/Cargo.toml:86-102

Configuration Profiles

Configuration is managed through Loco-native sections and MCB-specific settings:

• Development: Uses SQLite, Ollama embeddings, and Milvus.
• Test: Enables destructive test database flags and uses FastEmbed.
• Production: Requires admin API keys and utilizes optimized storage.

Sources: AGENTS.md:241-252, config/mcb-validate.toml:13-23

Admin Interface

Beyond the MCP protocol, mcb-server provides a web-based admin UI for monitoring and managing the server state. These routes are registered under the /ui prefix.

UI Controllers

The admin interface includes several functional pages:

• Dashboard: Overview of system status.
• Config: Displays active runtime configurations.
• Health: Shows system metrics and connectivity status.
• Jobs: Monitors background tasks like indexing.
• Browse: Interface for navigating stored code collections.

Sources: crates/mcb-server/src/controllers/web/mod.rs:133-140, README.md:65

Summary

Memory Context Browser (MCB) serves as a bridge between AI agents and complex codebases by providing a persistent, searchable, and structured memory layer. Its adherence to Clean Architecture and its modular provider system allow it to scale from local development environments to production-grade Kubernetes deployments.

Sources: README.md:4-21, AGENTS.md:154-162