MCP Tools Reference
Knowledge Raven provides 6 MCP tools. Your AI agent automatically selects the right tool based on your request — you don’t need to call them manually.
How the Agent Uses These Tools
Knowledge Raven uses Agentic RAG with 3 query types. The agent selects the optimal retrieval strategy automatically:
| Query Type | Best For | Alpha |
|---|---|---|
| Precision | Specific questions — “What is our vacation policy?“ | 70% semantic + 30% keyword |
| Explorative | Broad topics — “What do we know about the onboarding process?“ | 30% semantic + 70% keyword |
| Discovery | Browsing — “What documents are available about the Q3 project?” | Metadata-only, no vector search |
search_knowledge_base
Precision search within a specific knowledge base.
Use when you have a specific question and know which knowledge base it belongs to. Returns the most relevant chunks with parent context.
Example user prompts:
- “Search my knowledge base for our customer refund policy”
- “Find the API authentication documentation in the Engineering KB”
Returns: Ranked chunks with parent context, source links, and relevance scores.
broad_search
Explorative search across all knowledge bases.
Use when the question is broad, exploratory, or could span multiple knowledge bases. Uses BM25-heavy search to find keyword matches across all available content.
Example user prompts:
- “What does our company know about GDPR compliance?”
- “Find everything related to our Q3 OKRs”
Returns: Results from all knowledge bases the user has access to, ranked by relevance.
fetch_document
Retrieve a specific document — preview, full content, or by chunks.
Three modes:
- preview — First ~500 words + document summary. Use to check if a document is relevant before fetching the full content.
- full — Complete document text. Use when full context is needed.
- chunks — Specific chunks by ID. Use for targeted retrieval after search.
Example user prompts:
- “Show me the full text of the onboarding guide”
- “Get a preview of the Q3 report”
Returns: Document content with source link for deep-linking back to the original.
list_knowledge_bases
Discover available knowledge bases.
Returns all knowledge bases the current user has access to, with names, descriptions, and document counts.
Example user prompts:
- “What knowledge bases do I have access to?”
- “List all available knowledge bases”
Returns: Knowledge base list with metadata (name, description, document count, connector types).
list_documents
Browse documents in a knowledge base with summaries.
Returns documents with auto-generated 2–3 sentence summaries (generated at ingestion by GPT nano). Useful for discovery without performing a search.
Example user prompts:
- “What documents are in the HR knowledge base?”
- “List all documents about the product roadmap”
Returns: Document list with summaries, file types, and source links.
get_document_metadata
Retrieve metadata without loading full content.
Returns title, source URL, file type, last synced date, and connector type — without the document body. Useful when the agent needs to verify a document exists or get its source link.
Example user prompts:
- “When was the employee handbook last updated?”
- “Where is the original source for document X?”
Returns: Metadata including source_link for deep-linking to the original document.
Source Deep-Linking
All search and fetch tools return a source_link field. When you ask your agent about a topic, it can provide clickable citations that open the original document at the cited passage — directly in Confluence, Notion, GitHub, or your Knowledge Raven dashboard.