Local MCP server delivering semantic code context for AI agents
tracedecay, from ScriptedAlchemy, is a local MCP server that provides semantic code intelligence for AI coding agents, designed to reduce redundant file scanning. It builds a repository-wide semantic graph and exposes more than 70 specialized MCP tools to answer queries about call relationships, impact radius, and anchored edits, which cuts LLM token consumption by narrowing requests to relevant snippets. Engineers and AI researchers who integrate agent assistants get project memory and metadata kept on their own machines.
What tasks can you actually use it for?
tracedecay functions as a project memory that AI agents query for specific developer tasks. It supports discovery and navigation via a local semantic graph, impact analysis to assess downstream changes, and AST-based anchored edits for precise code modifications. Typical agent workflows it supports include call-graph queries, PR context assembly, diff-aware changelog generation, and mapping tests to code paths. These tasks are exposed through the toolset of over 70 MCP tools.
How accurate are the insights it produces?
The tool produces structured, machine-queryable code metadata by extracting symbols with Rust tree-sitter extractors across 50+ languages, which grounds agent answers in parsed syntax rather than raw text. Anchored edits use AST rewriting and string replacement tools, and impact radius analysis reports downstream dependencies. Accuracy depends on extractor coverage for each language and the fidelity of parsed trees; the Rust core is intended to keep processing fast and reduce resource overhead while building and querying graphs.
Does it integrate with workflows and protect project data?
tracedecay stores project memory locally in a libSQL database and runs as a Rust binary on macOS (Apple Silicon), Linux, and Windows (via Scoop), which keeps indexing and metadata on the developer's machine. It integrates with Git workflows for PR and diff analysis, but it requires an MCP-compliant client or AI agent to function. The implementation and active maintenance by ScriptedAlchemy aim to fit into agent-driven development pipelines rather than replace existing CI or editor tooling.
Practical choice for teams using agent-driven code review and refactoring
tracedecay is a pragmatic option for developers and researchers who want agent-aware project context without moving code off-host; its design favors local analysis and machine-queryable metadata. Expect a short integration step to connect an MCP client and to validate AST-based edits in your CI workflow. Use the tool as a source of targeted context for agents, and verify suggested changes before merging into main branches.
