Changelog
All notable changes to the SYKE MCP Server.
- Dynamic version: Server version is now read from
package.json at runtime instead of being hardcoded in 4 places. MCP handshake, startup log, update check, and Smithery sandbox all report the correct version automatically.
- Public website copy cleanup: Removed stale commercial-plan language across marketing pages so the website reflects the open-source + BYOK model consistently.
- Cortex page cleanup: Removed outdated plan-specific CTA copy and aligned onboarding pages around local usage, sample output, and docs.
- File viewer fix: Files with
../ prefix in path no longer cause “Server returned non-JSON” errors. Switched from URL path to query parameter approach.
- Website version updated: Landing page hero and statusbar now show v1.8.10+.
- Version consistency fix: Resolved version drift across
package.json, index.ts, and website — all now report v1.8.2. Homepage hero tag and statusbar updated from v1.4.0.
- Website metadata sync: Cortex landing metadata and related public copy were aligned so social previews and navigation match the current feature positioning.
- Cortex page aria-label fix: Cleaned up stale accessibility labels and navigation text on the onboarding pages.
- Version Sync API: New
/api/version endpoint returns current version, latest npm version, and outdated status with 24-hour registry cache.
- Startup update check: MCP server now checks for updates on startup (non-blocking). Logs an update prompt when a newer version is available on npm.
- Dashboard update banner: Top-of-page banner appears when your local SYKE is outdated, with copy-ready upgrade command.
- GitHub source published: Full source code available at github.com/khalomsky/syke.
scan_project tool: AI-powered full-project scan generates a comprehensive onboarding document covering architecture, key files, and patterns.- Static analysis wording: "tree-sitter" references updated to "static analysis" across docs and landing page for clarity.
- AI key management: Added REMOVE button to the AI keys modal in the dashboard. Clear any configured API key with one click to disable AI analysis per provider.
- SCC Condensation + Topological Sort: Tarjan's algorithm detects all circular dependency clusters, condenses the graph into a clean DAG, and computes correct cascade levels. Replaces the old DFS cycle detection (capped at 10) with O(V+E) full coverage.
- Composite Risk Scoring (0–1): Five weighted signals replace the old 3-bucket threshold. Fan-in (30%), Stability Index (20%), Cyclomatic Complexity (20%), Cascade Depth (15%), PageRank (15%). Every file gets a continuous risk score that AI agents can use for threshold decisions.
- Git Change Coupling: Mines the last 500 commits to find files that frequently co-change without import relationships. Surfaces hidden dependencies (config files, stylesheets, schemas) that static analysis misses. Integrated into
analyze_impact and gate_build outputs.
- PageRank for File Importance: Power Iteration PageRank adapted for dependency graphs. Files imported by many important files rank higher than those imported by leaf nodes. Every file gets a rank position and percentile.
get_hub_files now sorts by PageRank instead of simple dependent count.
- Incremental Graph Updates: Single-file changes now re-parse only the changed file and update edges in place, instead of rebuilding the entire graph. SCC and PageRank recompute after edge changes in <100ms.
- Memoized BFS Queries: Impact analysis results are cached with a reverse-index for smart invalidation. Repeated queries return in O(1). 500-entry LRU cache with hit/miss diagnostics exposed in
refresh_graph output.
- Gemini API safeguards: Added 1.5s file change debounce, MD5 hash caching (skip duplicate analysis), rate limiting (10 calls/min), and a toggle endpoint (
POST /api/toggle-realtime-ai) to disable real-time AI during active development.
- All 5 algorithms run locally with zero AI token cost. AI semantic analysis is reserved for high-risk changes only.
- Settings Panel (MCP Dashboard): Manage AI provider keys (Gemini / OpenAI / Anthropic) directly from the web dashboard — no config file editing needed.
- Admin Dashboard expansion: Active device count, live session tracking, country breakdown, and additional admin utilities.
- Multi-language UI: Auto-detects Korean/English from system locale. Configurable via
~/.syke/config.json or SYKE_LANGUAGE env var.
- Hot-swap keys: Change an AI provider key at runtime without restarting the MCP server.
- Version consistency fix: Server constructor, startup log, and package.json now all report the same version.
- Dashboard Settings Panel: 42 real-time controls across 9 groups — nodes, links, particles, layer colors, scene, camera, arrows, animation, physics.
- Smart Search: Search nodes by name, layer, or keyword. Matching nodes highlight with non-matches dimmed. Camera auto-frames results. Clearing restores full graph view.
- Defensive API layer: Content-type validation and URI-encoded file paths on all dashboard API calls. Prevents silent parse errors.
- Language Plugin System: 8 languages with zero-config auto-detection. No setup required — SYKE detects your stack automatically from file extensions.
- New languages added: Python, Go, Rust, Java, C++, Ruby (joining existing TypeScript/JavaScript and Dart support).
- Unified static analysis: Cross-language dependency extraction using a unified plugin architecture, replacing the previous language-specific approach.
- Internal refactor:
src/languages/plugin.ts registry replaces the Dart-only src/dart/ module.
- Web Dashboard: 3D interactive dependency graph visualization at
syke.cloud/dashboard.
- Real-time file monitoring: Server-Sent Events (SSE) stream pushes live graph updates as your AI modifies files.
- Animated diff viewer: Visual representation of cascade changes in the dashboard.
- Gemini AI analysis integration: Connect your Gemini API key for semantic impact analysis.
- New MCP tool:
check_warnings — scan your entire codebase for potential issues.
- Initial release. Core MCP server available as
@syke1/mcp-server on npm.
- Core MCP tools:
gate_build, analyze_impact, check_safe, get_dependencies, get_hub_files, refresh_graph, ai_analyze.
- Dart language support via lakos for dependency graph construction.
- Basic dependency graph analysis: PASS / WARN / FAIL build gate, impact traversal, hub file detection.