Currently, using the mcp if we have to query the claude code or any other ai client, or for them to use the search_docs tool, the documents need to be pre-fed into the qdrant db.
which includes running -
npx tsx src/main.ts ingest <owner> <repo> // fetches both pulls/issues from that repo
which would fetch Documents into the Document[] (one document for one issue/pull) -> chunking the documents using the chunking strategy -> embedding using the gemini-embedding-001 -> storing to qdrant. (also creating per documents hash that is stored in ./storage for the llamaIndex to know which documents changed and need to be re-embedded).
Create a tool ingest_docs which would allow ai clients to fetch the repo issues/pulls on demand and push into the rag pipeline by its own and then call the search_docs tool dynamically to answer the query instead of replying that the document wasn't found in the qdrant db.
Currently, using the mcp if we have to query the claude code or any other ai client, or for them to use the
search_docstool, the documents need to be pre-fed into the qdrant db.which includes running -
which would fetch Documents into the
Document[](one document for one issue/pull) -> chunking the documents using the chunking strategy -> embedding using thegemini-embedding-001-> storing to qdrant. (also creating per documents hash that is stored in ./storage for the llamaIndex to know which documents changed and need to be re-embedded).Create a tool
ingest_docswhich would allow ai clients to fetch the repo issues/pulls on demand and push into the rag pipeline by its own and then call thesearch_docstool dynamically to answer the query instead of replying that the document wasn't found in the qdrant db.