Stellarion is built to work with Claude Code. Once configured, all 47 tools become available as stellarion_* invocations from Claude.
npm install -g @stellarion/stellarion-mcp
Save as ~/.claude.json (user-scope) or .mcp.json in your project root (project-scope).
{
"mcpServers": {
"stellarion": {
"command": "stellarion-mcp",
"args": [
"--workspace", "/absolute/path/to/your/project"
]
}
}
}
For multi-repo projects, pass --workspace multiple times — every repo lands in the same logical graph, and cross-file imports resolve across them.
Start Claude Code and ask:
What Stellarion tools do you have available?
You should see the 28 community tools (or 47 if your trial / Pro license is active).
For hardware-design teams. Pass --mode eda to filter the tool surface for EDA workflows:
scan_security, find_implementors, find_related_tests, cross_project_search (don't apply to HDL/Tcl)find_eda_calls, get_flow_graph, find_duplicate_constraintsEDA mode requires a Pro license or active 180-day trial.
{
"mcpServers": {
"stellarion": {
"command": "stellarion-mcp",
"args": [
"--workspace", "/absolute/path/to/your/eda/project",
"--mode", "eda"
]
}
}
}
The TCL/SDC/UPF parser ships with Stellarion and classifies every command at parse time into 40+ stage-aware categories (quartus_assignment, qsys_module, qsys_interface, quartus_timing, query_clock, read_sdc, …). On a Quartus codebase like OFS FIM Common, this surfaces ~1,900 EDA call sites that the generic call graph cannot.
The Stellarion repo ships 14 architect-level slash commands that compose the Stellarion tools into focused, single-question workflows. Drop them into ~/.claude/commands/ (user-scope) or <project>/.claude/commands/ (project-scope):
git clone https://github.com/stellarion-ai/stellarion.git /tmp/stellarion
mkdir -p ~/.claude/commands
cp /tmp/stellarion/examples/claude-code/commands/*.md ~/.claude/commands/
Then in Claude Code:
/architecture-review
/complexity-hotspots
/eda-flow-audit
/eda-drift-check
/refactor-roi-report
/quality-scorecard
See the Slash Commands guide for the full list and what each one does.
What does emit_ip_cfg do? Show me callers and tests.
Claude chains stellarion_symbol_search → stellarion_get_ai_context → stellarion_find_related_tests.
I'm reviewing this branch. What's the risk?
Claude calls stellarion_assess_change_risk on the changed-files list, then drills into HIGH/CRITICAL items with stellarion_analyze_impact and stellarion_get_callers.
/architecture-review
The slash command composes stellarion_codebase_health + stellarion_tech_debt_report + stellarion_dependency_drift + stellarion_scan_security (or EDA tools in EDA mode) + stellarion_api_surface into a single ship-readiness report.
/eda-drift-check
Calls stellarion_find_duplicate_constraints with onlyTrueConflicts: true and classifies the output as bugs / parameterization candidates / managed forks.
/refactor-roi-report 20
Builds a prioritized backlog with explicit ROI scoring (impact ÷ effort), categorized as 🔥 Quick win / 💪 High value / 🧱 Big rock. The argument controls how many top items to include.
For best semantic-search results, index the workspace once before heavy querying:
Index this workspace.
Claude calls stellarion_reindex_workspace. Subsequent calls are incremental.
Find complexity hotspots in src/api/ -- not the whole repo.
Specifying a path keeps reports focused.
> /complexity-hotspots 10
[table of 10 hotspots]
> Explain the top one in detail.
[get_ai_context drill-in]
> What would happen if I refactored it?
[analyze_impact]
| Init-opt key | CLI flag | Default | Description |
|---|---|---|---|
mode | --mode <profile> | general | general / eda / hybrid |
maxFiles | --max-files <n> | 5000 | Per-workspace file cap |
maxFileSizeKB | (n/a) | 1024 | Skip files larger than this |
excludePatterns | --exclude <dir> | sensible defaults | Glob patterns to skip |
embeddingModel | --embedding-model <m> | bge-small | bge-small (384d, fast) or jina-code-v2 (768d, more accurate) |
See Installation for the full list.