Code understanding: turn your repo into a queryable graph
Purpose #
The v4 researcher agent has a code-graph (cgb) slot for dependency-topology
and impact analysis — "what calls this?", "what breaks if I change this?",
"how does this module relate to the rest of the system?". Two open-source,
MIT-licensed tools fill that slot. Both run inside your AI CLI with your own
credentials — aidokit makes zero LLM calls and neither installs, runs, nor
proxies these tools (it only points you at them).
| Tool | What it is | How aidokit relates to it |
|---|---|---|
| graphify | An MCP server (PyPI graphifyy) |
In the MCP catalog — aidokit mcp add graphify |
| Understand-Anything | An AI-CLI plugin (Lum1104/Understand-Anything) |
Documented here — installed via your CLI's plugin system, not the catalog (ADR-0030) |
Quick steer: want a catalogued MCP server scoped per-role by aidokit? Use
graphify. Want an interactive dashboard + a broad set of /understand-*
slash commands? Use Understand-Anything. They are not mutually exclusive.
Option A — graphify (catalogued MCP server) #
graphify parses your code locally with tree-sitter (no API calls for code);
only non-code content (docs, PDFs, images, video transcripts) is sent to your
configured AI model. It exposes MCP tools like query_graph, get_neighbors,
shortest_path, and get_pr_impact.
Because it reads the whole project tree and sends non-code content off-host,
its catalog entry is securitySensitive: true and triggers: ['never'] —
strictly opt-in, never auto-suggested, and --yes does not bypass the
consent prompt.
Install #
# Prereq: Python >= 3.10 and uv (or pipx) on your host. aidokit detects but
# never installs prereqs (ADR-0008).
aidokit mcp add graphify
This runs the per-adapter recipe (claude-code): claude mcp add graphify -- uvx --from "graphifyy[mcp]" graphify-mcp.
⚠ Build the graph first (ADR-0030 §3, smoke-tested 2026-06-09). The graphify MCP server serves a pre-built graph; it does not build one on startup. Before enabling it, build
graphify-out/graph.jsonfrom your project root (run graphify's/graphifyskill, oruvx --from graphifyy graphify .). If the graph is missing, the server exits withGraph file not found: …/graphify-out/graph.json.The catalog command resolves to the
graphify-mcpexecutable from thegraphifyy[mcp]extra (equivalent to upstream'spython -m graphify.serve); it serves over stdio and defaults tographify-out/graph.json, so no path argument is needed. The earlieruvx graphifyy mcpform was incorrect —graphifyyis not an executable and there is nomcpsubcommand. See the graphify README for build options.
Option B — Understand-Anything (AI-CLI plugin) #
Understand-Anything builds a knowledge graph of every file, function, class, and
dependency via a multi-agent pipeline, plus an interactive dashboard. It exposes
slash commands: /understand, /understand-dashboard, /understand-chat,
/understand-diff, /understand-explain, /understand-onboard,
/understand-domain, /understand-knowledge.
It installs through your AI CLI's plugin system, not through aidokit mcp add
— that is why it is documented here rather than listed in the MCP catalog
(it has no claude mcp add shell recipe; see ADR-0030 for the catalog boundary).
Install #
# Claude Code (in-session)
/plugin marketplace add Lum1104/Understand-Anything
/plugin install understand-anything
# Copilot CLI
copilot plugin install Lum1104/Understand-Anything:understand-anything-plugin
# Cursor / VS Code + Copilot: auto-discovered via the plugin manifest.
# One-line installer (macOS/Linux):
curl -fsSL https://raw.githubusercontent.com/Lum1104/Understand-Anything/main/install.sh | bash
# Windows PowerShell:
iwr -useb https://raw.githubusercontent.com/Lum1104/Understand-Anything/main/install.ps1 | iex
Then run /understand in your CLI to build the graph, and /understand-chat to
query it.
- Repo: https://github.com/Lum1104/Understand-Anything (MIT)
Because it is not a catalog entry,
aidokit doctorandaidokit mcp listwill not report Understand-Anything — its lifecycle belongs to your AI CLI.
Which fills the researcher cgb slot? #
Either does. graphify is the entry the catalog ships for the slot (ADR-0015); Understand-Anything is an equally valid documented alternative. If you scope MCPs per role with aidokit, graphify integrates into that flow directly; if you prefer the dashboard/slash-command UX, install Understand-Anything as a plugin alongside your aidokit-scaffolded workflow.
Trust boundary (read once) #
- Both tools run in your CLI with your model credentials. aidokit never calls a model and never ships your code anywhere.
- graphify keeps code local; non-code content goes to your model. Treat any off-host content the way your org's data policy requires.
- Neither tool is auto-installed or auto-enabled. You opt in explicitly.