fix(init): honor --no-mcp and --no-hooks for Claude Code

[michaelxer]

Fixes #118.

Problem

graft init --no-mcp (and --no-hooks) only reached other-agent wiring. Claude Code still wrote .mcp.json and the hook shim/settings, even though the README says --no-mcp skips Claude Code MCP registration.

That's the path that burns extra LLM cache when Claude Code is pointed at a DeepSeek endpoint.

Fix

• runInit honors mcp / hooks the same way runHostsInit already did
• --dry-run / the picker drop those Claude paths
• a later wiring refresh replays the stamp so an upgrade does not put them back
• default graft init is unchanged

How I tested

node --import tsx --test test/claude-init.test.ts test/claude-settings-merge.test.ts test/hosts-init.test.ts test/hosts-plan.test.ts test/upkeep.test.ts

80 passed.

Local repro after the change:

graft init <tmp> --no-build --agents claude --no-mcp --no-hooks

writes statusline + skill, no .mcp.json, no graft-hooks.cjs. Default --agents claude still writes both.

Did not run a live Claude Code session against DeepSeek.

[Christian-B-Koller]

Appreciate your work 👍

[github-actions]

🌱 graft blast radius

3 areas changed → 1 area can be affected. 3 dependent symbols, depth 2.
Tests: no test reaches CLI Wiring; 2 areas updated their tests.
Tag: @anirudhkumar-nanonets — 4 of 4 areas · @shhdwi — 3 of 4 areas

flowchart TB
  A0(("MCP Retraction<br/>3 symbols"))
  classDef reached fill:#D9EDF3,stroke:#3AA7C9,stroke-width:1.5px,color:#0E313C;
  class A0 reached;

Loading

Can be affected	Symbols	Nearest hop	Reached from
MCP Retraction	3	src/hosts/retract.ts:L348-L467 targets — calls, depth 1	Claude Initialization, Claude Global Setup

Who knows this code — 2 people across 4 areas

Area	Who knows it
Claude Global Setup · changed	@anirudhkumar-nanonets — 5 commits, last today · @shhdwi — 3 commits, last 19d ago
Claude Initialization · changed	@anirudhkumar-nanonets — 13 commits, last today · @shhdwi — 3 commits, last 1mo ago
CLI Wiring · changed	@anirudhkumar-nanonets — 35 commits, last today · @shhdwi — 23 commits, last 19d ago
MCP Retraction · affected	@anirudhkumar-nanonets — 2 commits, last today

Ownership is git history over each area's own files, weighted towards recent work (120-day half-life). Merge commits and bots are dropped, and you are dropped from your own PR. A name with no @ has no GitHub handle in its commit email — tag them by hand, or add a .mailmap entry. A suggestion from history, not a CODEOWNERS rule.

All 3 dependent symbols, grouped by area

MCP Retraction — 3 symbols in 1 file

• src/hosts/retract.ts:L348-L467 — targets (calls, depth 1)
  369: for (const t of claudeTargets(repo)) keptPaths.add(t.path);
• src/hosts/retract.ts:L474-L476 — planRetract (calls, depth 2)
• src/hosts/retract.ts:L483-L486 — runRetract (calls, depth 2)

Test signal per changed area — 2 ✓ · 1 ✗

Reached = a node under a test path has a resolved edge into the changed symbol. It undercounts anything called indirectly — through a CLI, a spawned process or a dynamic import — so read a low ratio as “look here”, never as a coverage gate.

• ✓ Claude Global Setup — 3 of 3 reached · 2 test files changed here: test/hosts-claude-global.test.ts, test/hosts-plan.test.ts
• ✓ Claude Initialization — 2 of 3 reached · 3 test files changed here: test/claude-init.test.ts, test/claude-settings-merge.test.ts, test/hosts-claude-global.test.ts
  • not reached: claudeTargets
• ✗ CLI Wiring — 0 of 2 reached · no test file reaches it
  • not reached: wireTarget, rewriteWiring

2 test suites also reference this code

4 symbols, kept out of the diagram and the table so they cannot crowd out the areas a reviewer has to look at.

• test/cli-picker.test.ts
• test/hosts-retract.test.ts

⚠️ 1 changed file not in the graph (CHANGELOG.md) — no parser claims the extension, or the index predates the file.

_{graft blast · origin/main...HEAD · depth 2 · 12 changed files}

Open the interactive graph → — click an area to see its dependent symbols at file:line.

[shhdwi]

Thanks for this, and for the clear write-up of #118! Main has moved underneath it: the Claude Code home wiring from #276 changed the MCP result types, and after resolving the conflicts this branch no longer compiles (src/hosts/claude-global.ts: 'McpWrite' is not assignable to 'ConfigWrite'). #333 is the one we are taking forward for the dry-run half. The real-init bug is still there on main (--no-mcp --no-hooks still writes .mcp.json and the hook shim for Claude Code), so a fresh, focused PR against current main for #118 would be very welcome. Closing this one.