# CCC & CocoIndex Integration — Codebase Semantic Search
How CocoIndex Code (CCC) and the CocoIndex framework provide semantic code search and data transformation pipelines for the Kings' College Galway project.
## CCC — CocoIndex Code
CCC provides semantic code search over the entire polyglot monorepo. It uses CocoIndex flows to build a vector index of all source files, enabling agents to search by meaning rather than text matching.
### Current Setup
```bash
# Check if index exists
ls -la .cocoindex_code/target_sqlite.db # 35+ MB index database
# Index the codebase (incremental — only changed files)
bun run ccc:index
# Search by meaning
bun run ccc:search "Dagster asset partition definition"
bun run ccc:search "Pangolin private resource Traefik label"
bun run ccc:search "BAML extraction function"
```
### MCP Server
```json
{
"mcp": {
"cocoindex-code": {
"type": "local",
"command": ["ccc", "mcp"]
}
}
}
```
This exposes the `cocoindex-code_search` tool to all agents, which is the **primary code discovery tool** — per the AGENTS.md instruction: "always use ccc before grep/find."
### Index Update Policy
```bash
# Manual incremental refresh (fast — only indexes changed files)
bun run ccc:index
# Full rebuild (slow — re-indexes entire codebase)
bun run ccc:init # Clears existing index
bun run ccc:index # Full build from scratch
# Scheduled re-index (via turbo.json)
bun run turbo ccc:index
```
### Search API
```python
# Programmatic search via CCC CLI
import subprocess
def ccc_search(query: str, limit: int = 5) -> list[dict]:
result = subprocess.run(
["bun", "run", "ccc:search", query, "--limit", str(limit)],
capture_output=True, text=True
)
return json.loads(result.stdout)
```
### Per-Agent Indexing
Each agent can trigger re-indexing for its domain before searching, ensuring fresh results:
```
# Agent workflow:
1. ccc:index # Refresh index (incremental, fast)
2. ccc:search "..." # Search with current index
```
## CocoIndex Flow Examples
The `docs/data_engineering/cocoindex/` directory contains 25+ example flows:
### Text Embedding with LanceDB
From `cocoindex/text_embedding_lancedb/`:
```python
from cocoindex import Flow, TextEmbedding, LanceDB
flow = Flow("text_embedding")
.source(TextEmbedding.from_markdown("./docs/**/*.md"))
.transform(LanceDB("curriculum_embeddings"))
flow.run()
```
### Document Knowledge Graph
From `cocoindex/docs_to_knowledge_graph/`:
```python
flow = Flow("docs_kg")
.source(DocumentSource("./docs/**/*.md"))
.transform(LLMExtraction(
model="deepseek-v4-pro",
schema=KnowledgeGraphSchema,
))
.sink(Neo4jSink("bolt://localhost:7687"))
```
### Code Embedding
From `cocoindex/code_embedding/`:
```python
flow = Flow("code_index")
.source(CodeSource("./**/*.py", chunker="tree-sitter"))
.transform(TextEmbedding("code-embeddings"))
.sink(LanceDBSink("./.cocoindex_code"))
```
### Multi-Format Indexing with ColPali
From `cocoindex/multi_format_indexing/`:
```python
flow = Flow("visual_docs")
.source(MultiFormatSource("./docs/**/*.{pdf,png,jpg}"))
.transform(ColPaliEmbedding())
.sink(QdrantSink("http://localhost:6333"))
```
## CCC + Cognee Workflow
```
┌─────────────────┐ ┌─────────────────┐
│ CCC │ │ Cognee │
│ (code search) │ │ (docs cognition) │
├─────────────────┤ ├─────────────────┤
│ Source: code │ │ Source: .md docs │
│ Index: SQLite │ │ Graph: Neo4j │
│ Search: vector │ │ Search: GraphRAG │
│ MCP: ccc mcp │ │ MCP: cognee-mcp │
└────────┬────────┘ └────────┬────────┘
│ │
└───────────┬───────────┘
│
┌────────▼────────┐
│ Agent Query │
│ "find how BAML │
│ extraction is │
│ implemented" │
└────────┬────────┘
│
┌───────────┼───────────┐
▼ ▼
CCC returns Cognee returns
code files documentation
(implementation) (architecture, patterns)
```
## Index File Structure
The `.cocoindex_code/` directory:
```
.cocoindex_code/
├── cocoindex.db # Flow state (incremental tracking)
├── settings.yml # Index configuration
└── target_sqlite.db # Semantic search index (~35 MB)
# Contains vector embeddings for all source files
# Queryable via ccc search or SQLite
```
## Performance Notes
| Operation | Scope | Time |
|:--|:--|:--|
| `ccc:index` (incremental) | Changed files only | <10s |
| `ccc:index` (full rebuild) | Entire monorepo | ~2-5 min |
| `ccc:search` | Semantic query | <1s |
| `ccc mcp` server | Continuous | Background |
## Related Documentation
- **CocoIndex Flows**: `docs/data_engineering/cocoindex/` — 25 example flows
- **CocoIndex API Research**: `docs/data_engineering/cocoindex/cocoindex-api-research.md`
- **CCC Skill**: `.agents/skills/ccc/SKILL.md` — Agent usage guide
- **AGENTS.md**: Root instruction — "always use ccc before grep/find"