Vectora
Delegate RAG, Tavily web search, filesystem work, and persistent memory to a dedicated MCP sub-agent while you ship features in Claude Code.
Overview
Vectora is an MCP server for the Build phase that acts as a RAG, web-search, filesystem, and memory sub-agent for Claude Code via the vectora-agent stdio package.
What is this MCP server?
- MCP sub-agent combining RAG, web search, filesystem access, and memory
- stdio PyPI package vectora-agent (version 0.1.0rc3)
- Requires Cohere (embeddings/rerank) and Tavily (search/URL extraction) API keys
- Optional LLM providers: Google Gemini, OpenAI, Anthropic via environment secrets
- Explicitly positioned as a sub-agent for Claude Code workflows
- PyPI identifier vectora-agent
- Four capability areas: RAG, web search, filesystem, memory
What problem does it solve?
Agent builders juggle separate RAG, search, file, and memory setups instead of one MCP sub-agent their main session can call.
Who is it for?
Solo developers already paying for Cohere and Tavily who want a packaged sub-agent during Claude Code feature work.
Skip if: Beginners avoiding API keys, teams that forbid filesystem tools, or products that need a hosted multi-tenant assistant without self-managed secrets.
What do I get? / Deliverables
After configuration, Claude Code can offload retrieval, live web context, local files, and remembered facts to Vectora through standard MCP tools.
- Running MCP sub-agent with RAG, search, filesystem, and memory tools
- Delegated research and context layers callable from the primary coding agent session
Recommended MCP Servers
Journey fit
Canonical shelf is Build because Vectora is agent infrastructure you wire in while constructing AI-assisted products. Agent-tooling is the right subphase: it extends the coding agent with retrieval, search, files, and memory rather than shipping end-user UI.
How it compares
Composite agent MCP sub-agent, not a single-purpose database or browser automation server.
Common Questions / FAQ
Who is Vectora for?
Vectora is for Claude Code and MCP users building agent-heavy workflows who want RAG, search, files, and memory behind one sub-agent server.
When should I use Vectora?
Use it while building agent features—docs Q&A, research copilots, or memory-backed coding assistants—once you have Cohere and Tavily keys ready.
How do I add Vectora to my agent?
Install vectora-agent from PyPI, set COHERE_API_KEY and TAVILY_API_KEY (plus optional LLM keys), register the stdio MCP server in Claude Code, and invoke it as a sub-agent per upstream instructions.