cortex-local 2.2.0__tar.gz → 2.2.2__tar.gz

{cortex_local-2.2.0 → cortex_local-2.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cortex-local
-Version: 2.2.0
+Version: 2.2.2
 Summary: Local semantic search MCP server — notes + code, embedded Qdrant, fastembed
 Requires-Python: >=3.11
 Requires-Dist: click>=8.0.0

{cortex_local-2.2.0 → cortex_local-2.2.2}/cortex/core/chunker.py

@@ -7,14 +7,14 @@ from pathlib import Path
 
 REPOS_DIR = Path(os.environ.get("REPOS_DIR", "/tmp/repos"))
 
-CODE_EXTS = {".js", ".ts", ".tsx", ".jsx", ".svelte", ".py", ".java", ".go", ".rs", ".css", ".html", ".kt", ".kts", ".gd", ".yml", ".yaml"}
-SKIP_DIRS = {"node_modules", ".git", "dist", "build", ".next", ".svelte-kit", "__pycache__", ".gradle", "target"}
+CODE_EXTS = {".js", ".ts", ".tsx", ".jsx", ".svelte", ".py", ".java", ".go", ".rs", ".css", ".html", ".kt", ".kts", ".gd", ".yml", ".yaml", ".cs"}
+SKIP_DIRS = {"node_modules", ".git", "dist", "build", ".next", ".svelte-kit", "__pycache__", ".gradle", "target", "bin", "obj"}
 LANG_MAP = {
     ".js": "javascript", ".ts": "typescript", ".tsx": "typescript",
     ".jsx": "javascript", ".svelte": "svelte", ".py": "python",
     ".java": "java", ".go": "go", ".rs": "rust", ".css": "css", ".html": "html",
     ".kt": "kotlin", ".kts": "kotlin", ".gd": "gdscript",
-    ".yml": "yaml", ".yaml": "yaml",
+    ".yml": "yaml", ".yaml": "yaml", ".cs": "csharp",
 }
 
 def _is_excluded(rel: str, patterns: list) -> bool:
@@ -97,6 +97,15 @@ _FRAMEWORK_IMPORTS = [
     ("use actix",              "actix"),
     ("use axum",               "axum"),
     ("use tokio",              "tokio"),
+    # C#
+    ("using Microsoft.AspNetCore",       "aspnetcore"),
+    ("using Microsoft.Extensions",       "aspnetcore"),
+    ("using Microsoft.EntityFrameworkCore", "entityframework"),
+    ("using Newtonsoft.Json",            "newtonsoft"),
+    ("using System.Text.Json",           "system-text-json"),
+    ("using Xunit",                      "xunit"),
+    ("using NUnit.Framework",            "nunit"),
+    ("using Microsoft.Azure",            "azure-sdk"),
 ]
 
 try:
@@ -124,6 +133,15 @@ try:
         ".kt":  {"function_declaration", "class_declaration", "object_declaration"},
         ".kts": {"function_declaration", "class_declaration", "object_declaration"},
     }
+    try:
+        import tree_sitter_c_sharp as _tscs
+        _TS_LANGUAGES[".cs"] = Language(_tscs.language())
+        _TS_SEMANTIC[".cs"] = {
+            "method_declaration", "class_declaration", "interface_declaration",
+            "constructor_declaration", "struct_declaration", "enum_declaration",
+        }
+    except Exception:
+        pass
     _TS_AVAILABLE = True
 except Exception:
     _TS_AVAILABLE = False

cortex_local-2.2.2/cortex/core/formatters.py

@@ -0,0 +1,114 @@
+"""Shared result formatters for MCP search tools — used by server and local modes."""
+
+from datetime import datetime, timezone
+
+
+def format_notes_parts(results) -> tuple:
+    """Build parts list from note search results.
+
+    Returns (parts, matched_files, matched_scores). Caller is responsible for:
+    - computing ppr_block from matched_files/matched_scores and appending to parts
+    - joining parts with '\\n\\n---\\n\\n'
+    - updating stats['context_tokens_notes']
+    """
+    parts = []
+    matched_files = []
+    matched_scores = []
+    for r in results:
+        p = r.payload
+        tags = p.get("tags", [])
+        tag_str = f" `{'`, `'.join(tags)}`" if tags else ""
+        parts.append(
+            f"**{p['file']} > {p['heading']}** (score: {round(r.score, 3)}){tag_str}\n\n{p['text']}"
+        )
+        matched_files.append(p.get("file", ""))
+        matched_scores.append(r.score)
+    return parts, matched_files, matched_scores
+
+
+def format_code_results(deduped) -> str:
+    """Build formatted string from scored/deduped code search results.
+
+    Input: list of (boosted_score, qdrant_point, file_meta) tuples.
+    Caller is responsible for updating stats['context_tokens_code'].
+    """
+    parts = []
+    for boosted_score, r, file_meta in deduped:
+        p = r.payload
+        centrality = file_meta.get("centrality")
+        community = file_meta.get("community_id")
+        header = (
+            f"**{p['repo']}/{p['file']}** "
+            f"lines {p['start_line']}-{p['end_line']} "
+            f"({p.get('language', '')}) — score: {round(boosted_score, 3)}"
+        )
+        if centrality is not None:
+            header += f" · centrality: {centrality}"
+        if community is not None:
+            header += f" · community: {community}"
+        url = p.get("github_url", "")
+        parts.append(f"{header}\n{url}\n\n{p['text']}" if url else f"{header}\n\n{p['text']}")
+    return "\n\n---\n\n".join(parts)
+
+
+def fmt_stats(v: str, stats: dict, current: bool = False) -> str:
+    """Format stats block for one version. Safe for both server and local stats dicts."""
+    total_searches = stats.get("search_code_calls", 0) + stats.get("search_notes_calls", 0)
+    ppr_rate = (
+        f"{stats['ppr_fires'] / stats['search_notes_calls'] * 100:.1f}%"
+        if stats.get("search_notes_calls", 0) > 0 else "—"
+    )
+    avg_ppr = round(stats["ppr_results_added"] / stats["ppr_fires"], 1) if stats.get("ppr_fires", 0) > 0 else "—"
+    total_cache = stats.get("graph_cache_hits", 0) + stats.get("graph_cache_misses", 0)
+    cache_rate = f"{stats['graph_cache_hits'] / total_cache * 100:.1f}%" if total_cache > 0 else "—"
+    avg_lift = round(stats["centrality_lift_total"] / stats["centrality_lift_count"], 4) if stats.get("centrality_lift_count", 0) > 0 else 0
+    total_results = stats.get("total_results_code", 0) + stats.get("total_results_notes", 0)
+    avg_results = round(total_results / total_searches, 1) if total_searches > 0 else "—"
+    notes_pct = f"{stats['search_notes_calls'] / total_searches * 100:.0f}%" if total_searches > 0 else "—"
+
+    label = f"{v} (current)" if current else v
+    lines = [label]
+
+    started = stats.get("started_at", "")
+    if current and started:
+        try:
+            delta = datetime.now(timezone.utc) - datetime.fromisoformat(started)
+            h, rem = divmod(int(delta.total_seconds()), 3600)
+            m = rem // 60
+            lines.append(f"  uptime: {h}h {m}m")
+        except Exception:
+            pass
+    elif started:
+        last = stats.get("last_saved", "")
+        period = started[:10]
+        if last:
+            period += f" -> {last[:10]}"
+        lines.append(f"  period: {period}")
+
+    lines += [
+        f"  searches: code={stats.get('search_code_calls', 0)} notes={stats.get('search_notes_calls', 0)}  total={total_searches}  (notes {notes_pct})",
+        f"  results: code={stats.get('total_results_code', 0)} notes={stats.get('total_results_notes', 0)}  avg/search={avg_results}",
+        f"  centrality lift avg: {avg_lift} (across {stats.get('centrality_lift_count', 0)} results)",
+        f"  PPR: {stats.get('ppr_fires', 0)} fires ({ppr_rate} of note searches) · +{stats.get('ppr_results_added', 0)} results · avg {avg_ppr}/fire",
+    ]
+    _ppr_diag = {k: stats.get(k, 0) for k in ("ppr_nx_missing", "ppr_graph_missing", "ppr_no_matches", "ppr_below_threshold", "ppr_exception")}
+    if any(_ppr_diag.values()):
+        lines.append(
+            f"  PPR misses: nx={_ppr_diag['ppr_nx_missing']} graph={_ppr_diag['ppr_graph_missing']}"
+            f" no_match={_ppr_diag['ppr_no_matches']} below_threshold={_ppr_diag['ppr_below_threshold']}"
+            f" exception={_ppr_diag['ppr_exception']}"
+        )
+    ctx_notes = stats.get("context_tokens_notes", 0)
+    ctx_code = stats.get("context_tokens_code", 0)
+    ctx_total = ctx_notes + ctx_code
+
+    def _k(n):
+        return f"{n // 1000}k" if n >= 1000 else str(n)
+
+    lines += [
+        f"  context injected: notes={_k(ctx_notes)} code={_k(ctx_code)}  total={_k(ctx_total)} tokens (~chars/4 est.)",
+        f"  graph cache: {cache_rate} ({stats.get('graph_cache_hits', 0)}/{total_cache})",
+        f"  embed cache skipped: notes={stats.get('embed_cache_notes', 0)} code={stats.get('embed_cache_code', 0)}",
+        f"  reindexes: {stats.get('reindex_count', 0)}",
+    ]
+    return "\n".join(lines)

{cortex_local-2.2.0 → cortex_local-2.2.2}/cortex/mcp_server.py

@@ -3,7 +3,6 @@
 import json
 import threading
 import time
-from datetime import datetime, timezone
 from pathlib import Path
 
 
@@ -19,6 +18,7 @@ from .config import (
 from .embedder import embed, sparse_embed
 from .state import _stats, get_graph_meta, save_stats, invalidate_graph_cache
 from .indexer import index_path
+from .core.formatters import format_notes_parts, format_code_results, fmt_stats
 
 mcp = FastMCP("cortex")
 
@@ -107,16 +107,7 @@ def _search_notes(query: str, limit: int = 5):
     if not results:
         return "No results found.", [], []
     _stats["total_results_notes"] += len(results)
-    parts = []
-    matched_files = []
-    matched_scores = []
-    for r in results:
-        p = r.payload
-        tags = p.get("tags", [])
-        tag_str = f" `{'`, `'.join(tags)}`" if tags else ""
-        parts.append(f"**{p['file']} > {p['heading']}** (score: {round(r.score, 3)}){tag_str}\n\n{p['text']}")
-        matched_files.append(p.get("file", ""))
-        matched_scores.append(r.score)
+    parts, matched_files, matched_scores = format_notes_parts(results)
     ppr = _ppr_block(matched_files, matched_scores)
     if ppr:
         parts.append(ppr)
@@ -170,23 +161,7 @@ def _search_code(query: str, limit: int = 5):
         {"repo": r.payload.get("repo", ""), "file": r.payload.get("file", ""), "score": round(bs, 3)}
         for bs, r, _ in deduped
     ]
-    parts = []
-    for boosted_score, r, file_meta in deduped:
-        p = r.payload
-        centrality = file_meta.get("centrality")
-        community = file_meta.get("community_id")
-        header = (
-            f"**{p['repo']}/{p['file']}** "
-            f"lines {p['start_line']}-{p['end_line']} "
-            f"({p.get('language', '')}) — score: {round(boosted_score, 3)}"
-        )
-        if centrality is not None:
-            header += f" · centrality: {centrality}"
-        if community is not None:
-            header += f" · community: {community}"
-        url = p.get("github_url", "")
-        parts.append(f"{header}\n{url}\n\n{p['text']}" if url else f"{header}\n\n{p['text']}")
-    out = "\n\n---\n\n".join(parts)
+    out = format_code_results(deduped)
     _stats["context_tokens_code"] += len(out) // 4
     return out, log_entries
 
@@ -290,34 +265,9 @@ def get_stats(all: bool = False) -> str:
             return "No persisted stats found."
         blocks = []
         for v in sorted(versions):
-            s = versions[v]
-            blocks.append(_fmt_stats(v, s, current=(v == VERSION)))
+            blocks.append(fmt_stats(v, versions[v], current=(v == VERSION)))
         return "\n\n".join(blocks)
-    return _fmt_stats(VERSION, _stats, current=True)
-
-
-def _fmt_stats(v: str, s: dict, current: bool = False) -> str:
-    total = s.get("search_code_calls", 0) + s.get("search_notes_calls", 0)
-    ppr_rate = f"{s['ppr_fires'] / s['search_notes_calls'] * 100:.1f}%" if s.get("search_notes_calls") else "—"
-    lift_avg = round(s["centrality_lift_total"] / s["centrality_lift_count"], 4) if s.get("centrality_lift_count") else 0
-    label = f"{v} (current)" if current else v
-    lines = [label]
-    started = s.get("started_at", "")
-    if current and started:
-        try:
-            delta = datetime.now(timezone.utc) - datetime.fromisoformat(started)
-            h, rem = divmod(int(delta.total_seconds()), 3600)
-            m = rem // 60
-            lines.append(f"  uptime: {h}h {m}m")
-        except Exception:
-            pass
-    lines += [
-        f"  searches: code={s.get('search_code_calls', 0)} notes={s.get('search_notes_calls', 0)}  total={total}",
-        f"  centrality lift avg: {lift_avg} (across {s.get('centrality_lift_count', 0)} results)",
-        f"  PPR: {s.get('ppr_fires', 0)} fires ({ppr_rate} of note searches) · +{s.get('ppr_results_added', 0)} results",
-        f"  reindexes: {s.get('reindex_count', 0)}",
-    ]
-    return "\n".join(lines)
+    return fmt_stats(VERSION, _stats, current=True)
 
 
 @mcp.tool()