VerveContext

Ground Claude in real-world truth

Claude natively speaks MCP, so grounding it is a config edit — not an integration project. Add VerveContext as an MCP server in Claude Desktop and Claude's answers start citing 300+ live, verifiable sources instead of guessing from training memory.

  • Claude Desktop
  • MCP-native
  • Cited answers
Get grounded→All frameworks
Why ground Claude

What connecting changes

Claude keeps its reasoning — it gains a live channel to facts, and a source on every claim.

Claude cites its sources

With VerveContext connected, Claude doesn't just state a number — it pulls the value from a real source and surfaces where it came from. A market briefing or research answer arrives with receipts a reviewer can follow.

  • Provenance
  • Traceable

Current, not remembered

Claude's training has a cutoff; the world doesn't. When a prompt needs today's gold price or the weather right now, Claude calls the source live instead of reaching for a stale figure from memory.

  • Live pull
  • No cutoff

Scoped access

Give a Claude project a key scoped to only the sources it should touch. The same Claude Desktop install can run a markets assistant and a research assistant on different, least-privilege keys.

  • Scoped keys
  • Per project
Setup

Add VerveContext to Claude Desktop

Claude reads MCP servers from claude_desktop_config.json — add one entry pointing at the endpoint with your scoped key, and restart.

  1. 01

    Open Claude Desktop's config

    Claude Desktop reads MCP servers from its claude_desktop_config.json. Open it from Settings → Developer → Edit Config, where you'll find (or create) an mcpServers object.

  2. 02

    Add VerveContext as an MCP server

    Inside mcpServers, add a "vervecontext" entry with "url": "https://api.vervecontext.com/v1/mcp" and a headers block carrying "x-api-key": "vc_live_…" — your scoped key.

  3. 03

    Restart Claude Desktop

    Claude reloads its MCP servers on restart. The sources you enabled appear as tools Claude can call — you'll see them listed in the tools menu of a new conversation.

  4. 04

    Ask something that needs a real number

    Ask Claude for a live figure and watch it call the source rather than guess. The answer comes back with the source attached — grounded, and ready to cite.

Grounding Claude, answered.

The Claude Desktop specifics teams ask about first.

MCP setup guides →
Does Claude support MCP natively?
Yes. Claude Desktop is an MCP client out of the box — that's the whole reason connecting VerveContext is a config edit rather than an integration project. You add one server entry and the sources become tools Claude can call.
Where does the config file live?
Claude Desktop exposes it under Settings → Developer → Edit Config, which opens claude_desktop_config.json. You add VerveContext under the mcpServers object there, then restart the app.
Will Claude call a source on every message?
No. Claude only calls a grounding tool when the answer actually needs a real value it doesn't hold. For anything it can answer from reasoning alone, it responds directly — the sources are there for the facts, not for every turn.
Can I limit which sources Claude can reach?
Yes — scope the key. A key granted only markets and weather means Claude simply won't have any other source available as a tool, and every call it does make is recorded in your usage logs.
What shows on my bill?
VerveContext runs on APIVerve, our production data engine, so invoices and card statements read APIVERVE. Same account, same key, same rails — nothing else changes.

Give Claude something to stand on. One MCP server entry and every answer comes back with its source.

Get grounded→How grounding works

Other agents?

Cursor, LangChain, or any MCP client connect the same way.

All connect guides