Connect Claude Code

SeldonFrame ships with an MCP server. Paste your token into Claude Code and it can read your CRM, build pages, ship agents — all from natural language.

What you'll get

Once Claude Code is connected, you can say things like:

  • "Build me a website chatbot for my HVAC business that can book appointments."
  • "Add John Smith (john@acme.com) to my CRM with stage 'qualified.'"
  • "Run evals on my customer-support agent and publish if ≥90% pass."
  • "Show me my agent's last 10 conversations and flag any that asked for a refund."
1

Get your MCP token

In the dashboard, open Settings → Integrations and copy your MCP bearer token. (Treat it like a password — it scopes Claude Code to your workspace.)
2

Install the SeldonFrame MCP server

It's an npm package. In any Claude Code project:
claude mcp add seldonframe \
  --transport http \
  --url https://app.seldonframe.com/api/mcp \
  --header "Authorization: Bearer <your-token>"
Or use the local stdio server (no token, scoped to a single workspace via env var):
npm i -g @seldonframe/mcp
claude mcp add seldonframe --command "npx @seldonframe/mcp"
3

Verify

In Claude Code, type /mcp and confirm seldonframe shows as connected with a list of tools. Then ask "What workspace am I connected to?" — Claude Code will call get_workspace_state and answer.

One tool to rule them all

SeldonFrame's MCP exposes 140+ tools, but Claude Code only needs two to get oriented: get_workspace_state (snapshot of your workspace) and build_website_chatbot (bundled "create + configure + run evals + publish" flow). We designed it that way deliberately — fewer tool calls, less thrashing, faster builds.

What the MCP can do

The full toolset covers:

  • CRM — read/write contacts, deals, bookings, custom fields.
  • Agents — create chatbots, update their personality, run evals, publish to live, embed on your site.
  • Pages — build landing pages, intake forms, booking pages.
  • Email — send transactional email, manage templates.
  • Automations — create rules that fire on CRM events.
  • Observability — read agent conversations, run telemetry queries, inspect eval failures.

BYOK for the LLM

Your agent's brain calls the LLM with your own Anthropic or OpenAI key, which you paste into Settings → LLM keys. SeldonFrame never holds the LLM bill — you do.

Next