Context+
An MCP server that transforms codebases into searchable, hierarchical feature graphs by combining Tree-sitter AST parsing (43 languages), spectral clustering, and Obsidian-style wikilink navigation. Provides 11 MCP tools for structural analysis, semantic code search, blast radius tracing, static analysis, and safe code modification with shadow restore points.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Context augmentation MCP. Enriches agent context with external data. Validate data sources. Sensitive context data persisted locally needs protection.
⚡ Reliability
Best When
You are working with large, multi-language codebases and want AI agents to have deep structural understanding including AST parsing, semantic clustering, and safe modification capabilities with undo support.
Avoid When
You have a small project where grep/ripgrep suffices, or you cannot run Ollama locally for embeddings. The dependency on a local LLM runtime adds setup friction.
Use Cases
- • Understanding large unfamiliar codebases through structural and semantic search
- • Tracing impact of changes via blast radius analysis before modifying code
- • Navigating multi-language projects using feature hubs and wikilink graphs
- • Safe AI code modifications with automatic shadow restore points for undo
Not For
- • Production deployment without local Ollama instance (requires local embedding model)
- • Teams that don't want a .mcp_data/ cache directory in their projects
- • Users needing cloud-hosted semantic search (this is entirely local)
Interface
Authentication
Requires OLLAMA_API_KEY environment variable for Ollama Cloud embedding operations. No other auth needed - runs entirely locally via stdio.
Pricing
MIT licensed. Fully open source. Requires local Ollama instance (also free) for embeddings. No cloud dependency.
Agent Metadata
Known Gotchas
- ⚠ Requires local Ollama instance running with embedding model (nomic-embed-text default)
- ⚠ Creates .mcp_data/ directory in project root for embedding cache
- ⚠ Embed batch size clamped to 5-10 range - cannot be tuned beyond those bounds
- ⚠ Shadow restore points are separate from git history - two undo systems to track
- ⚠ propose_commit is the only way to write code through this tool - enforces strict validation
Alternatives
Full Evaluation Report
Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for Context+.
Scores are editorial opinions as of 2026-03-06.