feat(cli): expand MCP editor detection and refactor for testability

[drewlyton]

Description

Adds MCP auto-configuration support for 5 new editors and refactors the detection architecture to eliminate a class of test fragility caused by global mock collisions.

New editors: Antigravity, Cline CLI, Codex CLI, GitHub Copilot CLI, MCPorter

Problem: Detection functions used module-level imports (existsSync, execa, process.platform), so the only way to test them was through global mocks in a 1,400-line integration test. Editors with overlapping filesystem paths (e.g., Cline's config lives inside VS Code's directory) caused mock collisions — a test for editor A would accidentally trigger editor B's detection. Adding new editors kept breaking existing tests.

Solution: Introduce a DetectionEnv interface that each detect function receives as a parameter instead of importing side-effectful modules directly. This makes every detect function independently testable with its own isolated context, and eliminates cross-editor mock interference.

What to review

Production code (minimal surface area):

• editorConfigs.ts — DetectionEnv type, createDetectionEnv() factory, all 13 detect functions updated to accept the env parameter, doc URLs added for every editor config path
• detectAvailableEditors.ts — accepts optional DetectionEnv, passes it through to each config.detect(ctx) call
• knip.config.ts — removed ignoreBinaries (no longer needed since execa calls go through the execCommand abstraction)

Test code (bulk of the diff):

• editorConfigs.test.ts — 44 new unit tests covering all 13 detect functions with isolated DetectionEnv contexts, including the exact Cline/VS Code collision case
• configure.test.ts — replaced 21 hand-written per-editor integration tests with a table-driven approach (1,411 → 820 lines). Behavioral edge-case tests (token reuse, auth expiry, config merging, etc.) are unchanged.

Testing

• 44 new unit tests for detect functions (each using isolated DetectionEnv — no global mock collisions)
• Table-driven integration tests covering all 13 editors + MCPorter file-format variants
• 12 behavioral edge-case tests preserved unchanged
• All 172 affected tests pass locally (macOS), all 43 CI checks pass (Ubuntu + Windows, Node 20/22/24)
• Type check, lint, dep check, and build all clean

[claude]

Claude finished @drewlyton's task in 3m 27s —— View job

---

PR Review

• Read CLAUDE.md for repo guidelines
• Review changed files
• Check changeset
• Post review findings

---

Changeset

The summary is two sentences; the second describes an implementation detail rather than user-facing behavior. Per Sanity product copy conventions it should be one concise sentence for a developer audience.

Current:

Added MCP auto-configuration support for Antigravity, Cline CLI, Codex CLI, GitHub Copilot CLI, and MCPorter. Refactored editor detection to use dependency injection for improved testability and cross-platform reliability.

Suggested:

Added MCP auto-configuration support for Antigravity, Cline CLI, Codex CLI, GitHub Copilot CLI, and MCPorter.

Fix this →

---

detectGeminiCli — behavioral narrowing

editorConfigs.ts:128-130

The old detection checked whether ~/.gemini/ directory existed, then returned ~/.gemini/settings.json as the target path (creating it if absent). The new code checks whether ~/.gemini/settings.json already exists:

// Before: directory presence → create or update settings.json
const geminiDir = path.join(homeDir, '.gemini')
return existsSync(geminiDir) ? path.join(geminiDir, 'settings.json') : null

// After: file must already exist
const settingsPath = path.join(ctx.homedir, '.gemini/settings.json')
return ctx.existsSync(settingsPath) ? settingsPath : null

This fixes the Antigravity false-positive (documented in the comment), but it also means a fresh Gemini CLI install where settings.json doesn't exist yet will not be auto-detected. The old behavior would have created the file; the new behavior silently skips the editor. If Gemini CLI creates settings.json on first run this is fine in practice — but it's a silent detection regression worth confirming is intentional.

[github-actions]

📦 Bundle Stats — @sanity/cli

Compared against main (5a7dcbe7)

@sanity/cli

Metric	Value	vs main (5a7dcbe)
Internal (raw)	2.1 KB	-
Internal (gzip)	799 B	-
Bundled (raw)	10.94 MB	-
Bundled (gzip)	2.05 MB	-
Import time	842ms	+4ms, +0.5%

bin:sanity

Metric	Value	vs main (5a7dcbe)
Internal (raw)	975 B	-
Internal (gzip)	460 B	-
Bundled (raw)	9.84 MB	-
Bundled (gzip)	1.77 MB	-
Import time	2.02s	+8ms, +0.4%

🗺️ View treemap · Artifacts

Details

• Import time regressions over 10% are flagged with ⚠️
• Sizes shown as raw / gzip 🗜️. Internal bytes = own code only. Total bytes = with all dependencies. Import time = Node.js cold-start median.

📦 Bundle Stats — @sanity/cli-core

Compared against main (5a7dcbe7)

Metric	Value	vs main (5a7dcbe)
Internal (raw)	93.8 KB	-
Internal (gzip)	21.9 KB	-
Bundled (raw)	21.62 MB	-
Bundled (gzip)	3.42 MB	-
Import time	801ms	+4ms, +0.5%

🗺️ View treemap · Artifacts

Details

• Import time regressions over 10% are flagged with ⚠️
• Sizes shown as raw / gzip 🗜️. Internal bytes = own code only. Total bytes = with all dependencies. Import time = Node.js cold-start median.

📦 Bundle Stats — create-sanity

Compared against main (5a7dcbe7)

Metric	Value	vs main (5a7dcbe)
Internal (raw)	976 B	-
Internal (gzip)	507 B	-
Bundled (raw)	50.7 KB	-
Bundled (gzip)	12.6 KB	-
Import time	❌ ChildProcess denied: node	-

Details

• Import time regressions over 10% are flagged with ⚠️
• Sizes shown as raw / gzip 🗜️. Internal bytes = own code only. Total bytes = with all dependencies. Import time = Node.js cold-start median.

[github-actions]

Coverage Delta

File	Statements
packages/@sanity/cli/src/actions/mcp/detectAvailableEditors.ts	82.5% (+ 0.5%)
packages/@sanity/cli/src/actions/mcp/editorConfigs.ts	98.7% (+ 23.0%)
packages/@sanity/cli/src/commands/mcp/configure.ts	90.0% (±0%)

Comparing 3 changed files against main @ 5a7dcbe708e5f8cfa922461c64e2c5d873668bad

Overall Coverage

Metric	Coverage
Statements	83.2% (+ 0.2%)
Branches	73.3% (+ 0.3%)
Functions	83.0% (+ 0.1%)
Lines	83.6% (+ 0.2%)