Integrations¶
Hub: documentation home · Python API · Getting started.
Two ways to use a tools.yaml. Pick the one your product supports.
| You are building… | Use | Install |
|---|---|---|
| A Python agent (LangChain, LangGraph, OpenAI Agents, Anthropic SDK) | load_tools / connect in-process |
Store extra (qdrant / pgvector / … — vector stores) plus langchain / langgraph / openai-agents / anthropic as listed below |
| A chat / IDE host (Claude Desktop, Claude Code, Codex, Cursor) | vectorsmith serve as an MCP server |
Store extra so vectorsmith is on PATH |
| A remote / Kubernetes MCP server | serve --http (self-host) |
Store extra plus auth-jwt / otel as needed |
Same YAML either way. Do not import Engine. Field-by-field: tools.yaml. Extras and public imports: library surface.
~50s — YAML tools, then the same file over MCP or in your Python SDK.
MCP hosts¶
- Claude Desktop
- Claude Code
- OpenAI Codex (CLI, IDE extension, ChatGPT desktop Codex)
- Cursor
- claude.ai custom connector (
serve --http)
Copy-paste configs: examples/mcp_hosts/.
Stdio remains on the handshake protocol even when a newer host probes
server/discover before initialize, covering the GitHub Copilot CLI
compatibility case. HTTP clients receive JSON-RPC -32022 with the supported
version list when their initialize version is unavailable. Details:
CLI protocol behavior.
Agent frameworks¶
- LangChain —
from vectorsmith import load_tools - LangGraph — same tools;
create_react_agent/ToolNode - OpenAI Agents SDK —
from vectorsmith.openai_agents import load_tools - Anthropic Messages API —
from vectorsmith.anthropic import load_tools - Other frameworks — LlamaIndex, CrewAI, raw
connect()