Skip to content

v1.0.0 — Production-ready: 12 MCP tools, Claude Code + Codex

Latest
Latest

Choose a tag to compare

View all tags
@lucianostraga lucianostraga released this 24 May 16:01
· 17 commits to main since this release
v1.0.0
02abadf

What's in v1.0.0

All 12 MCP tools verified end-to-end in both Claude Code and Codex against a live Agentforce Service Agent on a fresh scratch org.

Discovery

  • list_orgs — sf-authenticated Salesforce orgs
  • list_agents — Agentforce agents in an org
  • get_agent_metadata — full subagent + action map
  • load_config — read user expectations from sfagent-config.yaml

Live conversation

  • start_session / send_message / end_session — multi-turn via sf agent preview

Diagnostics (sf CLI 2026-05-20+)

  • list_traces / read_trace — read Salesforce's own session traces (with graceful fallback when CLI's parser dies)

Regression hand-off

  • generate_test_spec — emit YAML for sf agent test run-eval
  • run_batch_test / get_test_results — drive native test suites

Install

Claude Code:

claude plugin install sfagent-tools@sfagent-tools-marketplace

OpenAI Codex:

codex mcp add sfagent-tools -- npx -y sfagent-tools-mcp-server@latest

Any other MCP client (Cursor, Continue, Cline, Windsurf):

npx -y sfagent-tools-mcp-server@latest

What changed since the demo-video tag (Apr 8)

Restructured to a monorepo (packages/server + packages/claude-code-plugin + packages/codex-plugin). Aligned terminology with Agent Script v2.0 (subagent instead of topic). Added native trace tools and test-spec generation. Switched to npm-based distribution so any MCP client gets one-line install. Fully spec-compliant against both Claude Code and Codex official plugin docs.

Assets 2
Loading