Your AI is only as good as the context you give it.
General-purpose AI models lack the specific context your work demands. Kbrain.io plugs directly into your prompts - delivering curated, human expertise exactly when your AI needs it.
What happens when your AI doesn't know enough
General-purpose LLMs are impressive, but niche requirements, proprietary methodologies, and domain-specific frameworks often cause hallucinations or weak answers.
Fine-tuning and custom RAG can solve this, but both require engineering and ongoing maintenance.
KBrain is the third option: subscribe to an expert knowledge base and plug it into your assistant in minutes.
How it works for AI builders and assistants
Step 1 - Find the right brain
Browse the KBrain marketplace by category, topic, or use case.
Step 2 - Subscribe and get your API key
Subscribe monthly and receive secure credentials tied to your subscribed brain.
Step 3 - Plug it into MCP
Connect KBrain through MCP in Claude and start querying expert knowledge immediately. See the connection guide for step-by-step setup.
What you can plug KBrain into
- Claude (Anthropic) - Add KBrain as an MCP server for tool-based retrieval.
- Current scope - MCP workflows tested with Claude.
- Not yet supported - Non-MCP integrations and untested assistant platforms.
Available MCP tools
The KBrain connector exposes two read-only tools to every connected assistant:
list_brains- Returns the brains your account is actively subscribed to, along with asubscriptions_versionfingerprint.search_knowledge- Retrieves relevant knowledge from your subscribed brains. Accepts an optionalbrain_idto target one brain, or omit it to search across all subscribed brains.
Some curated brains expose additional domain-specific tools when you are subscribed to them - for example, the UNESCO UIS brain adds nine read-only tools for querying global education and literacy indicators. All KBrain MCP tools are read-only and never modify data in external systems.
What makes KBrain different from other knowledge sources
Human-curated, not scraped. Every brain is published by a domain expert.
Retrieval-optimized for AI. KBrain indexes for semantic relevance, not document storage only.
Access-controlled. You subscribe and access via keys; brains are not publicly indexed.
No infrastructure to maintain. No vector DB, embedding pipelines, or re-index cron jobs required.
Use cases
- Security and compliance agents
- Research assistants
- Customer-facing AI tools
- Internal knowledge management
- Developer tools and coding agents
Frequently asked questions
What is a KBrain brain?
A structured, retrieval-optimized knowledge base published by a domain expert and queryable via API.
How is this different from giving my agent a PDF?
A brain is semantically indexed and returns relevant excerpts on demand.
What AI tools does KBrain work with?
Today, KBrain supports MCP workflows tested with Claude. Broader integrations are not available yet.
How is billing handled?
Monthly subscription per brain. Cancel anytime.
Is the knowledge I query kept private?
Yes. Queries are private and not shared with other subscribers.
What if I need a brain that doesn't exist yet?
Request one on the marketplace or publish a private brain for your own team.
Troubleshooting
"No active brain connections for this user"
Your account has no active subscriptions. Subscribe to at least one brain from the marketplace, then retry.
401 Unauthorized / "OAuth bearer token required"
Your access token has expired or been revoked. In Claude, open Settings → Connectors, disconnect KBrain, and add it again to refresh the OAuth flow.
Tool call returns an empty result
Confirm the brain you queried is published and indexed. Run list_brains first to verify which brains are visible to the current session.
The assistant doesn't see a new subscription
The subscriptions_version fingerprint refreshes on every list_brains call. Ask the assistant to re-run list_brains - no reconnect required.
Still stuck? Reach out via our contact page.
Stop sending your AI in blind.
Kbrain.io is the context layer your prompts have been missing.