Add it once to every client you run

VectorHub is a remote MCP server over streamable HTTP, so any compliant client works. Here is the exact block to paste for each of the ones teams use most.

Get a key Full quickstart

Claude

Desktop client

Add VectorHub as a remote MCP server in Claude Desktop or Claude Code and every enabled connector shows up as callable tools.

bash

claude mcp add vectorhub \
  --transport http \
  https://mcp.vector-hub.org/v1 \
  --header "Authorization: Bearer $VECTORHUB_KEY"

Codex

CLI

toml

[mcp_servers.vectorhub]
url = "https://mcp.vector-hub.org/v1"
headers = { Authorization = "Bearer ${VECTORHUB_KEY}" }

Cursor

IDE

Drop the server into .cursor/mcp.json to give in-editor agents access to your issue tracker, warehouse and deploy tooling.

json

{
  "mcpServers": {
    "vectorhub": {
      "url": "https://mcp.vector-hub.org/v1",
      "headers": { "Authorization": "Bearer ${VECTORHUB_KEY}" }
    }
  }
}

Windsurf

IDE

Same JSON block, same key. Windsurf picks up tool changes without a restart.

json

{
  "mcpServers": {
    "vectorhub": {
      "serverUrl": "https://mcp.vector-hub.org/v1",
      "headers": { "Authorization": "Bearer ${VECTORHUB_KEY}" }
    }
  }
}

LangChain

Framework

Load VectorHub tools straight into a LangGraph agent through the MCP adapter.

python

from langchain_mcp_adapters.client import MultiServerMCPClient

client = MultiServerMCPClient({
    "vectorhub": {
        "transport": "streamable_http",
        "url": "https://mcp.vector-hub.org/v1",
        "headers": {"Authorization": f"Bearer {VECTORHUB_KEY}"},
    }
})
tools = await client.get_tools()

Anthropic SDK

Runtime

Pass the gateway as an MCP server on the Messages API and let the model call tools directly.

python

from anthropic import Anthropic

client = Anthropic()
message = client.beta.messages.create(
    model="claude-opus-5",
    max_tokens=2048,
    messages=[{"role": "user", "content": "Summarise this week's open deals"}],
    mcp_servers=[{
        "type": "url",
        "url": "https://mcp.vector-hub.org/v1",
        "name": "vectorhub",
        "authorization_token": VECTORHUB_KEY,
    }],
)

Anything else

The generic configuration

Every client integration reduces to the same three values. If yours speaks MCP over HTTP, this is all it needs.

mcp.json

{
  "mcpServers": {
    "vectorhub": {
      "url": "https://mcp.vector-hub.org/v1",
      "headers": {
        "Authorization": "Bearer ${VECTORHUB_KEY}"
      }
    }
  }
}

URL

https://mcp.vector-hub.org/v1

Transport

Streamable HTTP (SSE supported)

Auth

Bearer token: your scoped key

FAQ

Integration questions

Do I need a different key for each client?

You do not have to, but you should. A key per client, or per agent, is what makes the audit log useful and lets you revoke one without disturbing the others.

My client is not listed. Will it work?

If it speaks MCP over streamable HTTP, yes. The configuration is always the same three things: the URL, the Authorization header and a name.

Does the client need to be restarted when I add a connector?

No. The tool list is resolved per session at the gateway, so a connector enabled today is callable on the next conversation without touching client configuration.

Can I use VectorHub from a local stdio client?

Yes, through the CLI bridge: vectorhub mcp stdio proxies a local stdio transport to the remote gateway, which is useful for development and for clients that do not yet speak HTTP.

One endpoint, every client

Mint a key per agent and paste the same three lines wherever you run them.

Get early access Read the docs

No credit card required · Free tier available · Self-host on request