Capable

The CRM that lives in Claude. Capable ships a Model Context Protocol server so reps work their CRM by talking to an LLM — no second tab, no fields to fill out, no dashboards to maintain. Every read and write of customer data flows through Claude via MCP.

What it does

Paste a call transcript, email, or notes into Claude. Capable's propose_updates returns a list of atomic proposed changes (touches, field updates, SPICED fills, stage transitions, new contacts). You approve the ones that look right — field by field — and Claude commits them. You were in Claude the whole time. You never opened a CRM.

Connect

Endpoint: https://crm.capable.run/api/mcp — Streamable HTTP, OAuth 2.1 + PKCE (Bearer PAT also supported for older clients).

Claude.ai / Claude Desktop (OAuth — recommended)

​json { "mcpServers": { "capable": { "url": "https://crm.capable.run/api/mcp" } } } ​

Point your client at the URL — it auto-discovers the OAuth server, registers via DCR, and walks you through consent. No token to paste.

Claude Code CLI

​bash claude mcp add --transport http capable https://crm.capable.run/api/mcp ​

Older clients (Bearer PAT)

Create a PAT in the app (Settings → You → Developer) and pass it:

​bash claude mcp add --transport http capable https://crm.capable.run/api/mcp \ --header "Authorization: Bearer cpbl_pat_YOUR_TOKEN" ​

Tools

Accounts, contacts, opportunities, subscriptions, touches, signals, tasks, and reports — driven through generic verbs (create_record, update_record, search_records, delete_record) plus job-shaped domain verbs (log_touch, renew_subscription, get_pipeline_summary, …) and the flagship propose_updates → review_proposed_updates approval flow. Every tool returns the affected record's full state and carries MCP annotations + output schemas. Full catalog: https://crm.capable.run/learn

Schema

Eight core objects designed around revenue motions (not 2003-CRM objects): Accounts, Contacts, Opportunities, Subscriptions, plus Touches, Signals, Tasks, and universal Notes. Admins can switch objects on/off or add custom ones — the agent operates on the workspace's own schema.