Set up Claude Desktop

Live

Ask Claude “Is this on-brand?” inside any chat — BCP runs as a Model Context Protocol server Claude Desktop can connect to.

How it works

BCP exposes a Streamable-HTTP MCP server at /mcp. Claude Desktop runs a small stdio shim (npx mcp-remote) that proxies your local Claude session to the BCP endpoint. Your API key travels on every call as an Authorization: Bearer header; the server filters tool visibility by the key’s permissions and forwards each call to the same backend the REST API uses.

What you can do

  • Check a support reply against your brand voice before sending
  • Rewrite a Slack draft to fix tone or forbidden words
  • Audit ad copy and product descriptions against policy

Try these prompts

Once connected, type these straight into Claude:

  1. /load-brand

    Pulls your brand's voice, vocabulary, and governance rules into the session, so everything after this is brand-aware.

  2. "Draft a LinkedIn post announcing our new sustainability report. Raw notes: [paste your bullet points]. Keep it on-brand."

    Claude writes it, then self-checks with evaluate_text and revises until it passes — you get the draft plus its on-brand score.

  3. "Draft a reply to a customer who received a poor-quality item from an unauthorized reseller — on-brand and within policy."

    Balances empathy with brand protection and stays clear of claims your guardrails don't allow.

  4. /policy-check "Audit this ad copy strictly for compliance — ignore tone. [paste copy]"

    Governance-only verdict: flags risky claims, forbidden tactics, and required disclaimers — no tone rewrite.

  5. "Rewrite this email in our brand voice and remove any forbidden words: [paste]"

    Returns a before/after with a change log explaining each edit.

Setup

Requires Node.js 18+ (the config runs npx, which ships with Node).

i
Check Node first: run node -v in a terminal. If it prints v18 or higher, you're set. If you see command not found or a lower version, Node isn't installed — Claude Desktop will fail to start the server with a generic error (not an obvious "Node missing" message). Install Node.js, then fully restart Claude Desktop.

Add this to your Claude Desktop config file:

{
  "mcpServers": {
    "bcp": {
      "command": "npx",
      "args": [
"-y", "mcp-remote",
"https://www.brandcontextprotocol.com/mcp",
"--header",
"Authorization:Bearer bcp_"
      ]
    }
  }
}

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json

Fully quit Claude Desktop (Cmd+Q) and reopen. Ask “What MCP tools do you have available?” to confirm.

MCP tools your assistant sees

When you connect Claude Desktop or ChatGPT via MCP, the assistant sees these tools. Visibility is filtered per request by your API key’s permissions — a read_bcp-only key sees only get_brand_context.

Tool Permission What it does
get_brand_context read_bcp Fetch the brand’s BCP, scoped to the key’s allowed sections
evaluate_text evaluate Score text against verbal voice and policy rules
rewrite_text evaluate Suggest on-brand rewrites with optional target style
evaluate_design evaluate Score structured design data (colors, fonts, logo placement)
analyze_design_image evaluate Score a rendered image (banner, ad, social post) via vision
evaluate_document evaluate Score a multi-page PDF or image set end-to-end
check_brand_compliance evaluate Smart dispatcher — picks text / image / document by input shape

What it can do with BCP

  • Retrieve brand context Verified by BCP
  • Retrieve product context and claims —
  • Retrieve channel policy —
  • Evaluate output Verified by BCP
  • Rewrite / remediate Verified by BCP
  • Trigger human approval —
  • Report execution —
  • Provide governance evidence —