ContextAtlas is an MCP (Model Context Protocol) server positioned to provide AI coding assistants with a context service that fuses code structure and architectural intent. According to its official description, it combines LSP-level structural information with architectural intent from Architecture Decision Records (ADRs), git history, and tests, thereby significantly reducing token usage in architecture-related queries while keeping answers consistent.
Main capabilities
- get_symbol_context: Obtain the complete context of a symbol in a single MCP call, including its type, callers, rationale from ADRs, and related tests.
- find_by_intent: Search based on what the code is for rather than just its name, using FTS5 BM25 under the hood to retrieve commit records, ADRs, and tests.
- impact_of_change: Before modifying a symbol, traverse its structural and historical blast radius to help assess the parts a change may affect.
How to use Official materials describe it as an MCP server that provides context bundles for Claude Code. This means it needs to be connected to a client that supports MCP, so that AI can call it when querying code architecture, symbol context, or impact analysis. As for the specific installation commands, configuration steps, supported client versions, or environment requirements, the materials do not currently provide them.
Use cases
- When you need AI to understand architectural decisions and design rationale in a large codebase, not just syntactic structure.
- When you want AI to propose or execute design choices based on the project's own design themes and scope (rather than generic best practices).
- When assessing the structural and historical blast radius before editing a symbol.
- When reducing token consumption for architecture-related queries; officially stated to be measured at a 45%–72% reduction in a locked 24-prompt benchmark on hono and httpx.
Limitations Official materials do not state the supported programming languages, repository size, integration methods, pricing model, or platform compatibility. The token reduction data is based on a specific benchmark (24 prompts on hono and httpx), and actual results may vary by project. Specific installation and configuration details should be found in the official documentation.
