mache, from Agentic Research, converts source code and structured data into a navigable graph to aid code comprehension and agentic tooling. The tool runs as a Model Context Protocol server and exposes a project topology of function call chains, type hierarchies, and cross-references that agents and humans can traverse. Key capabilities include graph-based exploration, schema inference, and semantic search powered by tree-sitter parsing. Target users are developers and AI researchers who need structured context for automated code analysis and navigation.

What tasks can you actually use it for?

The tool converts flat source files into an explorable topology so agents and engineers can inspect logical relationships instead of reading files linearly. Typical tasks include dependency tracing and architecture inspection, plus supplying structured context to agents. Example uses:

• Code comprehension across modules
• Tracing function call chains
• Feeding graph context into agent queries

How accurate is the graph representation?

Parsing accuracy depends on tree-sitter integration, which the tool uses to infer schemas and cross-references. That approach maps language constructs to graph nodes with close correspondence for supported languages such as Go and Python, reducing manual mapping. Complex semantic interpretations still require developer verification when agents act on graph-derived suggestions, so outputs should be treated as structured assistance rather than authoritative judgments.

What file types and languages does it accept?

The tool processes conventional source files and structured data via tree-sitter, explicitly supporting languages including Go and Python. It exposes repository metadata either as a mounted filesystem or through its MCP interface so MCP-aware clients receive the same graph view. Repositories composed of standard source files integrate directly; projects that embed proprietary or binary-only artifacts need prior extraction to source form before ingestion.

Is it easy to include in agentic workflows?

Deployment targets development environments and agent pipelines; the tool is written in Go and runs on Windows, macOS, and Linux, requiring a Go runtime and MCP-compatible clients. Installation uses the Go toolchain and the project repository as the package source. Integration is direct for teams that already use MCP-aware agents, but teams must configure clients and adapt agent prompts to navigate specific graph relations rather than raw file text.

Best for engineering teams embedding agents into code navigation

The developer's focus on bridging language models and structured technical data makes the tool suited to teams building agentic navigation into codebases. Treat agent outputs as draft recommendations that need human verification, and include a review step in any automated code-change process. Practical advice: restrict agent queries to explicit graph relations to narrow scope and reduce irrelevant or speculative responses.