elara-labs · rajkumarsakthivel · May 8, 2026 · May 7, 2026 · May 7, 2026 · May 7, 2026
@@ -151,7 +151,9 @@ With CCE:       context_search "payment flow"            =    800 tokens
 
 We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (53 source files, 180K tokens) with 20 real coding questions. No cherry-picking, no synthetic queries.
 
-**Methodology:** For each query, "without CCE" means reading the full content of every file the query touches. "With CCE" means the relevant chunks after compression. This is conservative (agents often read more files than needed).
+**Methodology:** For each query, "without CCE" means reading the full content of every file the query touches. "With CCE" means the relevant chunks after compression.
+
+**Important baseline note:** The 94% number is measured against full-file reads, not against what Claude Code actually does. In practice, Claude Code already uses grep, partial file reads, and targeted tools, so the real-world savings compared to normal Claude Code behavior will be lower than 94%. We use full-file as the baseline because it's reproducible and deterministic (no agent behavior variability). The benchmark measures CCE's retrieval efficiency, not a head-to-head comparison with Claude Code's built-in exploration.
 
 | Metric | Result |
 |--------|--------|

@@ -0,0 +1,186 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8">
+  <meta name="viewport" content="width=device-width, initial-scale=1.0">
+  <title>How to Use CCE with Cursor: Complete Setup Guide (2026)</title>
+  <meta name="description" content="Step-by-step guide to setting up Code Context Engine with Cursor. Reduce token costs, get faster responses, and keep cross-session memory. Works alongside Cursor's built-in indexing.">
+  <meta name="keywords" content="Cursor setup guide, CCE Cursor, Cursor reduce costs, Cursor MCP server, Cursor token savings, Cursor code indexing, Code Context Engine Cursor, Cursor AI optimization">
+  <meta property="og:title" content="How to Use CCE with Cursor: Complete Setup Guide">
+  <meta property="og:description" content="Save tokens and get cross-session memory in Cursor. One command setup, works alongside built-in indexing.">
+  <meta property="og:type" content="article">
+  <link rel="canonical" href="https://elara-labs.github.io/code-context-engine/blog/cce-cursor-setup-guide.html">
+  <link rel="icon" type="image/svg+xml" href="../logo.svg">
+  <link href="https://fonts.googleapis.com/css2?family=Instrument+Sans:wght@400;500;600;700;800&family=DM+Mono:wght@400;500&display=swap" rel="stylesheet">
+  <style>
+    *, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
+    :root { --bg:#06080f; --bg2:#0c1120; --bg3:#131c30; --cyan:#00d4ff; --green:#34d399; --amber:#fbbf24; --text:#e8f0ff; --text2:#7b93b8; --text3:#3a5070; --border:#1a2a42; --mono:'DM Mono',monospace; --sans:'Instrument Sans',system-ui,sans-serif; }
+    body { background:var(--bg); color:var(--text); font-family:var(--sans); line-height:1.7; }
+    nav { position:sticky; top:0; z-index:100; padding:16px 48px; background:rgba(6,8,15,.85); backdrop-filter:blur(12px); border-bottom:1px solid var(--border); display:flex; align-items:center; justify-content:space-between; }
+    nav a { color:var(--text2); text-decoration:none; font-size:14px; font-weight:500; }
+    nav a:hover { color:var(--cyan); }
+    .nav-brand { display:flex; align-items:center; gap:10px; color:var(--text); font-weight:700; font-size:15px; }
+    .nav-brand img { width:24px; height:24px; border-radius:5px; }
+    article { max-width:720px; margin:0 auto; padding:80px 24px 120px; }
+    .meta { font-size:13px; color:var(--text3); font-family:var(--mono); margin-bottom:32px; }
+    h1 { font-size:40px; font-weight:800; line-height:1.2; letter-spacing:-0.025em; margin-bottom:24px; }
+    .subtitle { font-size:20px; color:var(--text2); margin-bottom:56px; line-height:1.5; }
+    h2 { font-size:26px; font-weight:700; margin-top:56px; margin-bottom:20px; }
+    h3 { font-size:20px; font-weight:700; margin-top:36px; margin-bottom:14px; }
+    p { font-size:17px; color:var(--text2); margin-bottom:20px; }
+    strong { color:var(--text); }
+    ul, ol { margin:0 0 20px 24px; color:var(--text2); font-size:17px; }
+    li { margin-bottom:8px; }
+    code { font-family:var(--mono); font-size:.88em; background:var(--bg3); border:1px solid var(--border); border-radius:4px; padding:2px 6px; color:var(--cyan); }
+    pre { background:var(--bg2); border:1px solid var(--border); border-radius:10px; padding:24px; font-family:var(--mono); font-size:14px; color:var(--text); overflow-x:auto; margin:24px 0; line-height:1.8; }
+    .callout { background:var(--bg2); border-left:3px solid var(--cyan); border-radius:0 10px 10px 0; padding:20px 24px; margin:24px 0; }
+    .callout p { margin-bottom:0; }
+    .step { display:flex; gap:20px; margin:24px 0; align-items:flex-start; }
+    .step-num { width:36px; height:36px; border-radius:50%; background:var(--bg3); border:1px solid var(--border); display:flex; align-items:center; justify-content:center; font-family:var(--mono); font-size:14px; font-weight:700; color:var(--cyan); flex-shrink:0; }
+    .step-content { flex:1; }
+    .step-content p { margin-bottom:8px; }
+    table { width:100%; border-collapse:collapse; margin:24px 0; font-size:15px; }
+    th { text-align:left; padding:12px 16px; font-size:12px; font-weight:600; letter-spacing:.05em; text-transform:uppercase; color:var(--text3); border-bottom:2px solid var(--border); }
+    td { padding:12px 16px; border-bottom:1px solid var(--border); color:var(--text2); }
+    td:first-child { color:var(--text); font-weight:500; }
+    .highlight { color:var(--cyan); font-weight:700; }
+    .cta-box { background:var(--bg2); border:1px solid rgba(0,212,255,0.2); border-radius:12px; padding:36px; margin-top:56px; text-align:center; }
+    .cta-box h3 { font-size:22px; margin-bottom:12px; margin-top:0; }
+    .cta-box p { text-align:center; }
+    .cta-link { display:inline-block; margin-top:20px; color:var(--cyan); text-decoration:none; font-weight:600; font-size:16px; }
+    .cta-link:hover { text-decoration:underline; }
+    .tags { margin-top:56px; padding-top:24px; border-top:1px solid var(--border); display:flex; flex-wrap:wrap; gap:8px; }
+    .tag { font-family:var(--mono); font-size:12px; padding:4px 12px; background:var(--bg3); border:1px solid var(--border); border-radius:20px; color:var(--text3); }
+    footer { border-top:1px solid var(--border); padding:24px 48px; text-align:center; font-size:13px; color:var(--text3); }
+    footer a { color:var(--text2); text-decoration:none; }
+    footer a:hover { color:var(--cyan); }
+    @media (max-width:640px) { article { padding:48px 16px 80px; } h1 { font-size:28px; } nav { padding:14px 16px; } .step { flex-direction:column; gap:12px; } }
+  </style>
+  <script type="application/ld+json">
+  { "@context":"https://schema.org", "@type":"Article", "headline":"How to Use CCE with Cursor: Complete Setup Guide", "datePublished":"2026-05-07", "author":{"@type":"Organization","name":"Elara Labs"} }
+  </script>
+</head>
+<body>
+<nav>
+  <a href="../" class="nav-brand"><img src="../logo.svg" alt="CCE"> Code Context Engine</a>
+  <a href="https://github.com/elara-labs/code-context-engine">GitHub</a>
+</nav>
+<article>
+  <div class="meta">May 7, 2026 · 5 min read</div>
+  <h1>How to Use CCE with Cursor</h1>
+  <p class="subtitle">Set up Code Context Engine in Cursor. Works alongside built-in indexing. One command, no conflicts.</p>
+
+  <h2>Why add CCE to Cursor?</h2>
+  <p>Cursor has built-in code indexing. So why add CCE? Three reasons:</p>
+  <ul>
+    <li><strong>Cross-session memory.</strong> Cursor starts fresh every session. CCE remembers decisions, architectural context, and code areas across sessions.</li>
+    <li><strong>Token savings tracking.</strong> Cursor doesn't tell you how many tokens it used. CCE tracks every query with dollar amounts.</li>
+    <li><strong>Multi-editor portability.</strong> If you also use Claude Code, VS Code, or Gemini CLI, CCE's index follows you. Cursor's doesn't.</li>
+  </ul>
+
+  <div class="callout">
+    <p><strong>CCE does not replace Cursor's indexing.</strong> They run side by side. Cursor's built-in features (tab completion, inline edits, apply) still use its own index. CCE adds MCP-based semantic search on top.</p>
+  </div>
+
+  <h2>Setup (2 minutes)</h2>
+
+  <div class="step">
+    <span class="step-num">1</span>
+    <div class="step-content">
+      <p><strong>Install CCE</strong></p>
+      <pre>uv tool install code-context-engine</pre>
+      <p>Or <code>pipx install code-context-engine</code>. Requires Python 3.11+.</p>
+    </div>
+  </div>
+
+  <div class="step">
+    <span class="step-num">2</span>
+    <div class="step-content">
+      <p><strong>Initialize in your project</strong></p>
+      <pre>cd /path/to/your/project
+cce init</pre>
+      <p>CCE auto-detects Cursor and writes <code>.cursor/mcp.json</code> plus <code>.cursorrules</code> with search instructions.</p>
+    </div>
+  </div>
+
+  <div class="step">
+    <span class="step-num">3</span>
+    <div class="step-content">
+      <p><strong>Restart Cursor</strong></p>
+      <p>That's it. Cursor now has access to CCE's 9 MCP tools. The AI will use <code>context_search</code> automatically when exploring your code.</p>
+    </div>
+  </div>
+
+  <h2>What gets created</h2>
+  <table>
+    <tr><th>File</th><th>Purpose</th></tr>
+    <tr><td><code>.cursor/mcp.json</code></td><td>Registers CCE as an MCP server</td></tr>
+    <tr><td><code>.cursorrules</code></td><td>Tells Cursor's AI to use context_search instead of reading files</td></tr>
+    <tr><td><code>.gitignore</code></td><td>CCE entries added (local cache, settings)</td></tr>
+  </table>
+
+  <h2>Verify it works</h2>
+  <pre>cce search "your feature name"</pre>
+  <p>If results come back with token counts, CCE is working. In Cursor, ask the AI any question about your codebase. It should call <code>context_search</code> instead of reading entire files.</p>
+
+  <h2>Check your savings</h2>
+  <pre>cce savings</pre>
+  <p>Shows tokens saved, dollar amounts, and per-layer breakdown. Run after a few Cursor sessions to see the impact.</p>
+
+  <h2>What CCE gives Cursor's AI</h2>
+  <table>
+    <tr><th>Tool</th><th>What it does</th></tr>
+    <tr><td><code>context_search</code></td><td>Semantic search across your codebase</td></tr>
+    <tr><td><code>expand_chunk</code></td><td>Get full source for a compressed result</td></tr>
+    <tr><td><code>related_context</code></td><td>Find imports, callers, dependencies</td></tr>
+    <tr><td><code>session_recall</code></td><td>Recall past decisions</td></tr>
+    <tr><td><code>record_decision</code></td><td>Save decisions for next session</td></tr>
+  </table>
+
+  <h2>CCE vs Cursor indexing: what handles what</h2>
+  <table>
+    <tr><th>Feature</th><th>Cursor built-in</th><th>CCE</th></tr>
+    <tr><td>Tab completion</td><td class="highlight">Yes</td><td>No</td></tr>
+    <tr><td>Inline edits</td><td class="highlight">Yes</td><td>No</td></tr>
+    <tr><td>Semantic code search</td><td>Yes</td><td class="highlight">Yes (MCP)</td></tr>
+    <tr><td>Cross-session memory</td><td>No</td><td class="highlight">Yes</td></tr>
+    <tr><td>Token savings tracking</td><td>No</td><td class="highlight">Yes</td></tr>
+    <tr><td>Works in other editors</td><td>No</td><td class="highlight">Yes (6 editors)</td></tr>
+    <tr><td>Code stays local</td><td>No (cloud)</td><td class="highlight">Yes</td></tr>
+  </table>
+
+  <h2>Troubleshooting</h2>
+
+  <h3>Cursor doesn't use context_search</h3>
+  <p>Check that <code>.cursor/mcp.json</code> exists and Cursor was restarted after <code>cce init</code>. Cursor needs a restart to pick up new MCP servers.</p>
+
+  <h3>Index seems stale</h3>
+  <p>Git hooks reindex on every commit. If you haven't committed, run <code>cce index</code> manually. Or run <code>cce watch</code> in a separate terminal for real-time updates.</p>
+
+  <h3>Want to remove CCE</h3>
+  <pre>cce uninstall</pre>
+  <p>Removes all CCE files, config, hooks, and index data. Cursor's built-in indexing is unaffected.</p>
+
+  <div class="cta-box">
+    <h3>Try it now</h3>
+    <p>One command. Works alongside Cursor's built-in indexing. No conflicts.</p>
+    <pre>uv tool install code-context-engine && cce init</pre>
+    <a class="cta-link" href="https://github.com/elara-labs/code-context-engine">View on GitHub</a>
+  </div>
+
+  <div class="tags">
+    <span class="tag">#Cursor</span>
+    <span class="tag">#CursorAI</span>
+    <span class="tag">#MCPServer</span>
+    <span class="tag">#CodeContextEngine</span>
+    <span class="tag">#SaveTokens</span>
+    <span class="tag">#AICoding</span>
+    <span class="tag">#DeveloperTools</span>
+    <span class="tag">#CursorSetup</span>
+  </div>
+</article>
+<footer>
+  <a href="https://github.com/elara-labs/code-context-engine">Code Context Engine</a> · MIT License · <a href="../">Home</a>
+</footer>
+</body>
+</html>
@@ -24,6 +24,12 @@
     <changefreq>monthly</changefreq>
     <priority>0.8</priority>
   </url>
+  <url>
+    <loc>https://elara-labs.github.io/code-context-engine/blog/cce-cursor-setup-guide.html</loc>
+    <lastmod>2026-05-07</lastmod>
+    <changefreq>monthly</changefreq>
+    <priority>0.9</priority>
+  </url>
   <url>
     <loc>https://elara-labs.github.io/code-context-engine/blog/cce-vs-cursor-indexing.html</loc>
     <lastmod>2026-05-05</lastmod>

@@ -137,7 +137,8 @@ def prune_orphans(self, known_hashes: set[str]) -> int:
             for i in range(0, len(orphan_list), 500):
                 batch = orphan_list[i : i + 500]
                 placeholders = ",".join("?" * len(batch))
-                self._conn.execute(
+                # Safe: placeholders is only "?" chars; values are parameterized.
+                self._conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
                     f"DELETE FROM embedding_cache WHERE content_hash IN ({placeholders})",
                     batch,
                 )

@@ -403,9 +403,10 @@ def _write_vec_row(conn, table: str, rowid: int, vec) -> None:
     doesn't break inserts on the source table — the failed row simply won't
     be semantically searchable until the vec tables are rebuilt.
     """
+    # Safe: table name is an internal constant, never from user input.
     try:
-        conn.execute(f"DELETE FROM {table} WHERE rowid = ?", (rowid,))
-        conn.execute(
+        conn.execute(f"DELETE FROM {table} WHERE rowid = ?", (rowid,))  # nosemgrep: sqlalchemy-execute-raw-query
+        conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
             f"INSERT INTO {table}(rowid, embedding) VALUES (?, ?)",
             (rowid, _serialize_vec(vec)),
         )
@@ -786,17 +787,18 @@ def prune_old_rows(
 
     archived: dict[str, list[dict]] = {}
 
+    # Safe: table and col_list are internal constants, never from user input.
     def _harvest_and_delete(table: str, columns: list[str], cutoff: int) -> int:
         col_list = ", ".join(columns)
-        rows = conn.execute(
+        rows = conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
             f"SELECT {col_list} FROM {table} WHERE created_at_epoch < ?",
             (cutoff,),
         ).fetchall()
         if not rows:
             return 0
         if archive:
             archived[table] = [dict(r) for r in rows]
-        conn.execute(
+        conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
             f"DELETE FROM {table} WHERE created_at_epoch < ?",
             (cutoff,),
         )

@@ -84,7 +84,8 @@ def _delete_files_sync(self, file_paths: list[str]) -> None:
         with self._lock:
             for batch in batched_params(file_paths):
                 placeholders = ",".join("?" * len(batch))
-                self._conn.execute(
+                # Safe: placeholders is only "?" chars; values are parameterized.
+                self._conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
                     f"DELETE FROM chunks_fts WHERE file_path IN ({placeholders})",
                     batch,
                 )

@@ -162,22 +162,23 @@ def _sync_delete_by_files(self, file_paths: list[str]) -> None:
         with self._lock:
             cur = self._conn.cursor()
             # Collect node IDs in batches to respect SQLite param limits.
+            # Safe: ph is only "?" chars; values are parameterized.
             node_ids: list[str] = []
             for batch in batched_params(file_paths):
                 ph = ",".join("?" * len(batch))
-                cur.execute(
+                cur.execute(  # nosemgrep: sqlalchemy-execute-raw-query
                     f"SELECT id FROM nodes WHERE file_path IN ({ph})", batch
                 )
                 node_ids.extend(row[0] for row in cur.fetchall())
             # Delete edges and nodes in batches.
             for batch in batched_params(node_ids):
                 ph = ",".join("?" * len(batch))
-                cur.execute(
+                cur.execute(  # nosemgrep: sqlalchemy-execute-raw-query
                     f"DELETE FROM edges WHERE source_id IN ({ph}) "
                     f"OR target_id IN ({ph})",
                     batch + batch,
                 )
-                cur.execute(f"DELETE FROM nodes WHERE id IN ({ph})", batch)
+                cur.execute(f"DELETE FROM nodes WHERE id IN ({ph})", batch)  # nosemgrep: sqlalchemy-execute-raw-query
             self._conn.commit()
 
     # ------------------------------------------------------------------

@@ -112,10 +112,11 @@ def _ensure_vec_table(self, dim: int) -> None:
                 self._conn.execute("DROP TABLE IF EXISTS chunks_vec")
                 self._conn.execute("DELETE FROM chunks")
                 self._conn.execute("DELETE FROM chunk_compressions")
-            self._conn.execute(f"""
-                CREATE VIRTUAL TABLE IF NOT EXISTS chunks_vec
-                USING vec0(embedding float[{dim}])
-            """)
+            # Safe: dim is a validated integer, never from user input.
+            self._conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
+                f"CREATE VIRTUAL TABLE IF NOT EXISTS chunks_vec "
+                f"USING vec0(embedding float[{dim}])"
+            )
             self._dim = dim
             self._conn.commit()
 
@@ -242,20 +243,21 @@ async def delete_by_files(self, file_paths: list[str]) -> None:
         from context_engine.utils import batched_params
 
         with self._lock:
+            # Safe: placeholders is only "?" chars; values are parameterized.
             for batch in batched_params(file_paths):
                 placeholders = ",".join("?" * len(batch))
                 if self._dim is not None:
-                    self._conn.execute(
+                    self._conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
                         f"DELETE FROM chunks_vec "
                         f"WHERE rowid IN (SELECT rowid FROM chunks WHERE file_path IN ({placeholders}))",
                         batch,
                     )
-                self._conn.execute(
+                self._conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
                     f"DELETE FROM chunk_compressions "
                     f"WHERE chunk_id IN (SELECT id FROM chunks WHERE file_path IN ({placeholders}))",
                     batch,
                 )
-                self._conn.execute(
+                self._conn.execute(  # nosemgrep: sqlalchemy-execute-raw-query
                     f"DELETE FROM chunks WHERE file_path IN ({placeholders})",
                     batch,
                 )