Product1 publisher3 min readPublished
Salesforce exposes its platform to coding agents through more than 60 new MCP tools
Headless 360 makes capabilities that used to sit behind the Salesforce console callable as APIs, MCP tools and CLI commands, and more than 60 of the new tools point coding agents such as Claude Code and Cursor at live orgs.
The Product Desk · Product desk
What happened
- Salesforce introduced Headless 360, which exposes the capabilities it says agents need most as APIs, MCP tools and CLI commands so humans and agents can build and act on any surface.
- The release also includes an experience layer that renders native interactions on Slack, Voice and WhatsApp, plus tools for controlling how agents behave in production before and after launch.
- Salesforce says the DevOps Center MCP folds a build loop that spanned four tools into one connected experience and cuts cycle times by as much as 40%.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision Someone now has to decide which identity and scope carry those 90-plus calls before the first developer opens a coding agent against a production org, and that decision belongs to whoever owns the org.
- cost The labour this creates lands on admins and release managers: mapping each new tool to a permission, and answering later for a config change that was made from an editor Salesforce doesn't run.
- exposure Every developer with org credentials and a coding agent is now a path into live data and workflows, so the thing to inventory is the tool list.
The most concrete account of what a coding agent does with a live org sits in Salesforce's own announcement, in a customer quote. Adones Guerra, a tech lead at Grupo Globo, said Agentforce Vibes has been "especially valuable for speeding up day-to-day tasks like metadata updates and quick tweaks, reducing manual work and boosting efficiency" [5]. He said the team primarily uses it for repetitive, time-consuming work: config analysis, metadata adjustments, generating boilerplate code, tests, small refactors and pre-deployment checks [6].
The same post quotes Elia Wallen, CEO of Engine, who said "With Agentforce, we've been able to deploy sophisticated, production-ready AI agents in just 12 days, driving millions in savings while significantly increasing our technical velocity" [7]. Two customers, two different jobs. One shipped agents that handle service work; the other has an assistant editing configuration. The second is what most orgs will see first, and it is the one that touches metadata.
Parker Harris, a Salesforce co-founder, asked "Why should you ever log into Salesforce again?" last month, and the post presents the question as direction [10]. By the company's own description, 25 years of Salesforce meant a human working inside Salesforce, a rep opening a console and clicking into a case to update its status [11].
Count the new surface. More than 60 MCP tools plus more than 30 preconfigured coding skills is more than 90 named ways to reach the platform without opening it [8], and Salesforce describes that access as complete and live, covering all of an org's data, workflows and business logic [2]. The calls originate in editors Salesforce does not operate, among them Claude Code, Cursor, Codex and Windsurf [2].
Control is Salesforce's own third pillar, described as new tools that govern how agents behave in production, before launch and after [3]. Salesforce did not disclose how those controls line up with existing profiles and permission sets, or set pricing and availability dates [9].
Salesforce says the build loop that used to require context-switching across four tools now happens in one connected experience, cutting cycle times "by as much as 40%" [4]. That is a best case, not a measured average.
For whoever signs off on this, sorting by agent is the wrong axis, because the same tool list is reachable by every coding agent a developer installs. The list itself needs sorting twice: does the call read or write, and can a person undo the write in the UI without a deploy. Reads under a scoped user go first. Writes a person can revert by hand are a pilot. Writes that need a release to reverse keep a human on the approval. Guerra's list spans both ends, from config analysis to metadata adjustments [6].
What to watch
- Whether Salesforce publishes how the 60-plus MCP tools inherit profiles, permission sets and field-level security, and at which license tier.
- Whether the production control tools record which change came from an agent, and who inside the customer can read that record.
- Release notes or support threads showing what happens when an MCP tool writes metadata into a production org from a developer's editor.