Weaviate Claude Skills

Why use it

Key features

Five skills covering the full local Weaviate lifecycle
Docker-compose template for a local instance
Batch ingestion and file-based imports
RAG queries with filters, hybrid search, and analytics
Python-venv setup with requirements.txt

Live Demo

What it looks like in practice

weaviate-claude-skills.replay ▶ ready

0/0

Install

Author repository · Claude Skills documentation

Use Cases

Real-world ways to use Weaviate Claude Skills

Stand up a local RAG over your docs in under an hour

👤 Developers starting a local RAG prototype ⏱ ~60 min intermediate

When to use: You want to chat with your own docs without sending them to a cloud vendor.

Flow

Docker-start Weaviate

Use weaviate-local-setup to start docker-compose for a local Weaviate.✓ Copied

→ docker ps shows weaviate healthy on 8080
Create a collection

Create a collection 'Docs' with fields: title(text), body(text), source_url(text).✓ Copied

→ Schema applied; weaviate-collection-manager returns config
Ingest

Ingest every .md under ./docs as objects in 'Docs'.✓ Copied

→ Batch progress reported; objects visible via query
Ask

Use weaviate-query-agent: 'How do we rotate API keys?'✓ Copied

→ Answer cites source_url from the Docs collection

Outcome: A runnable local RAG you can iterate on.

Pitfalls

Ingesting giant PDFs whole without chunking — Chunk at ~500 tokens; let the skill's ingestion template handle it

Combine with: filesystem

Iterate a collection schema without manual API calls

👤 Teams refining their vector schema ⏱ ~20 min intermediate

When to use: Early RAG — your schema changes weekly and you don't want curl commands.

Flow

Inspect current schema

Show the schema for 'Docs' — fields, vectorizer, index config.✓ Copied

→ Full config returned
Propose a change

Add a 'language' enum property (en/zh/ja); keep the rest.✓ Copied

→ Migration plan + backup caveat
Apply

Apply; reindex the existing 500 objects.✓ Copied

→ Job progresses; counts reconcile

Outcome: Schema evolved safely with awareness of reindex cost.

Pitfalls

Assuming property removals are free — Some changes require re-embedding; the skill flags when it's cheap vs expensive

Hybrid search + filter for reliable answers

👤 Anyone shipping RAG to real users ⏱ ~15 min intermediate

When to use: Pure semantic search misses exact-match terms; hybrid (BM25 + vector) fixes it.

Flow

Run hybrid queries

Query 'SAML SSO error code 42', alpha=0.5 (balance semantic + keyword).✓ Copied

→ Top hits include the literal '42' match but also conceptually related ones
Layer filters

Filter language='en' and source_url contains '/docs/auth/'.✓ Copied

→ Hits narrowed; answer stays cited

Outcome: Higher precision on exact-term queries without losing recall.

Combine with: local-rag

Combinations

Pair with other MCPs for X10 leverage

weaviate-claude-skills + filesystem

Ingest local files via filesystem MCP, then query through the skill

List all markdown under ./docs, ingest into Weaviate 'Docs', then hybrid-search for the question.✓ Copied

weaviate-claude-skills + local-rag

Compare approaches — embeddings-only vs Weaviate's structured index

For this corpus, benchmark local-rag vs Weaviate hybrid on 10 eval questions.✓ Copied

weaviate-claude-skills + qdrant

Evaluate the vector DB choice on your workload

Run the same eval set on Weaviate and Qdrant; report latency + MRR.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call
weaviate-local-setup (SKILL)	(none)	First time only — brings up local Weaviate
weaviate-connection (SKILL)	URL + API key	Verify reachability and auth
weaviate-collection-manager (SKILL)	schema ops	Create, inspect, or evolve a collection
weaviate-data-ingestion (SKILL)	object(s) or file paths	Single insert, batch insert, or file import
weaviate-query-agent (SKILL)	query + filters + alpha	Semantic / hybrid search and RAG answers

Cost & Limits

What this costs to run

API quota: None — Weaviate runs locally
Tokens per call: Query + retrieved chunk size
Monetary: Free for local. Weaviate Cloud has its own pricing if you migrate later.
Tip: Chunk intentionally (~500 tokens); oversize chunks waste both storage and context.

Security

Permissions, secrets, blast radius

Credential storage: Weaviate API keys in .env, not in skill files. Local dev can skip auth but don't expose the port.

Data egress: Nothing leaves your machine for local setup. Cloud Weaviate sends to your configured endpoint.

Local Weaviate bound to 0.0.0.0 by default in many images — restrict to 127.0.0.1 if other services run on your LAN.
Embedding models: if using a third-party vectorizer, your chunks go to that provider. Keep it local (e.g., text2vec-transformers) for sensitive data.

Troubleshooting

Common errors and fixes

Connection refused 8080

Weaviate container not running. docker compose up -d in weaviate-local-setup.

Verify: curl localhost:8080/v1/.well-known/ready

Schema change fails

Some property changes require collection recreate. Export, drop, recreate, re-ingest.

Verify: weaviate-collection-manager diff output

Slow queries

Check HNSW index parameters and that vectorizer is properly configured. Avoid full-text fallback on unindexed fields.

Verify: Query explain in the skill's debug mode

Alternatives