Twenty CRM Skill

Deterministic bash wrappers for the Twenty CRM REST API. One script per endpoint (396 total). Each script self-documents its JSON body.

Quick start

# Discover operations
./scripts/twenty.sh list-ops | jq '.[].operation'

# Get an input schema for any operation
./scripts/twenty.sh schema create company

# Run an operation
./scripts/twenty.sh create company '{"name":"Acme Inc"}'
./scripts/twenty.sh list people
./scripts/twenty.sh get person <id>
./scripts/twenty.sh batch-create companies '[{"name":"A"},{"name":"B"}]'

Also callable as raw scripts (e.g. ./scripts/list-people.sh) for cron, CI, or agents that prefer direct invocation.

Setup

export TWENTY_BASE_URL="https://api.twenty.com"
export TWENTY_API_KEY="your-key"

Or drop a .env file at the repo root — scripts/check-env.sh auto-loads it without overriding already-exported values.

Discovery

references/manifest.json — machine-readable catalog: {operation, script, method, path, object, category} for every endpoint. Regenerate with scripts/gen-manifest.sh.
references/manifest.md — human-readable index grouped by action.
scripts/schema.sh <script> — JSON Schema for any script's input (inferred from its usage() text).

Agent harnesses

Claude Code / OpenClaw — this SKILL.md is read directly. Invoke as /twenty-crm or via natural language.
MCP clients (Claude Desktop, Cursor, Zed, Windsurf) — see mcp-server/README.md for a stdio server wrapping all 396 endpoints as tools.
Any other agent / CLI — read references/manifest.json, call scripts directly with $TWENTY_API_KEY + $TWENTY_BASE_URL in env.

Workflow notes

Always list-* with filters before create-* to avoid duplicates.
Run any create-*.sh or update-*.sh with no arguments to print its expected JSON body.
Path parameters (IDs) are passed as positional args. JSON bodies are a single argument.

twenty-crm