vector-collection-management

name	vector-collection-management
skill_type	skill
description	Create, populate, list, and delete vector-store collections through the vector-mcp MCP server's vector_collection_management tool. Use when the agent must stand up a new RAG collection, ingest documents (from a directory, file paths/URLs, or raw text) into an existing collection, enumerate collections, or drop one — across any supported backend (chromadb, postgres/pgvector, qdrant, couchbase, mongodb). Do NOT use for running queries against a collection (use vector-hybrid-search) or for choosing/tuning a backend connection (vector-backend-operations).
license	MIT
tags	["vector","rag","collections","embeddings","mcp"]
metadata	{"author":"Genius","version":"0.1.0"}

Vector Collection Management

Lifecycle operations on vector-store collections — the named indexes that hold embedded document chunks — via the vector-mcp MCP server. One condensed, action-routed tool covers create / add / list / delete so the same call shape works across every backend.

When to use

• Create (or get-or-create) a collection to hold a corpus for RAG.
• Ingest documents into a collection — from a document_directory, a list of document_paths (files/URLs), or inline document_contents strings.
• List the collections that exist on a backend.
• Delete a collection you no longer need.

When NOT to use

• Querying / retrieving from a collection → vector-hybrid-search.
• Picking a backend, wiring host/port/credentials, or comparing engines → vector-backend-operations.
• Persisting derived knowledge into the epistemic-graph KG → this is a vector backend, not a KG source; use the graph-os tools for KG writes.

Prerequisites & environment

Connect via the mcp-client skill against the vector-mcp MCP server. Every action takes the connection parameters inline (or falls back to server defaults):

Parameter	Notes
db_type	Backend: chromadb, postgres, qdrant, couchbase, mongodb
db_path	On-disk path (e.g. for embedded chromadb)
host / port	Networked backends
db_name / username / password	Auth for postgres / couchbase / mongodb
collection_name	Target collection (defaults to memory in the retriever)

Embedding is handled server-side by the configured embedding model (agent-utilities[embeddings-openai] or HuggingFace/sentence-transformers extras).

Tools & actions

Condensed tool	Actions
vector_collection_management	create_collection, add_documents, delete_collection, list_collections

Key parameters

• overwrite — on create_collection, replace an existing collection instead of get-or-create.
• document_directory / document_paths / document_contents — the three mutually-usable ingestion inputs for create_collection and add_documents.
• confirm — required guard on delete_collection.

Recipes

Create a collection and seed it from a directory:

{"action":"create_collection","db_type":"chromadb","db_path":"./chroma","collection_name":"handbook","document_directory":"./docs"}

Add inline text to an existing collection:

{"action":"add_documents","db_type":"chromadb","db_path":"./chroma","collection_name":"handbook","document_contents":["VPN setup: connect to gw-hq before syncing."]}

List collections on a backend:

{"action":"list_collections","db_type":"qdrant","host":"qdrant","port":"6333"}

Delete a collection (guarded):

{"action":"delete_collection","db_type":"chromadb","db_path":"./chroma","collection_name":"handbook","confirm":true}

Gotchas

• create_collection is get-or-create by default; pass overwrite:true to rebuild from scratch — this drops existing vectors.
• delete_collection needs confirm:true; without it the call is a no-op guard.
• Ingest inputs are supported one-at-a-time per call; loaders derive from llama-index-readers-file so PDFs/EPUBs/HTML are parsed, but very large directories should be chunked into batches to avoid long-running calls.
• The backend must be reachable with the given db_type/connection params or the call errors before any write.

Related

• vector-hybrid-search — retrieve from the collections you populate here.
• vector-backend-operations — backend selection and connection details.