Is MCP a security model?

No. MCP is how an agent reaches a tool. Identity, least privilege, and audit still sit in your IdP, Salesforce sharing, and the gateway.

Who keeps the controls current after go-live?

We can, on the Run plan: set hours a month, a weekly read of transcripts and errors, and a written monthly note. The plans and prices are on the What you get page.

Salesforce MVP Hall of Fame (Shivanath Devinarayanan) · Founded 2022

Controls on the agents you actually ship.

Regime mapping is the floor. Policy lives on Flex Gateway: identity, field, row, action. Semantic checks at the edge, not in the prompt.

Get the written assessment Flex Gateway

Founded 2022 · Founder in the Salesforce MVP Hall of Fame · Dubai HQ · Kochi studio

Regimes

Mapped to the agents, not to a poster.

Poster or control

A policy document, and a control.

Governance on a poster

Governance on the hop

A slide that names the EU AI Act: A risk tier per agent, from the work it does
One shared service key for every agent: On-behalf-of identity per caller
No list of MCP servers in use: Every connector and approver named in a register
Transcripts nobody reads: A person reviews the exceptions on a set day

Gateway

Four policy families and a semantic check. On the hop.

If these only exist in a document, you do not have governance. You have a poster.

01
Identity
On-behalf-of · No shared service account · Business group per caller
02
Field
Which attributes leave the system · Masking at the gateway
03
Row
Which records this identity may touch · Sharing still applies
04
Action
Read · Write · Refuse · Escalate to a person
05
Semantic check
PII detection · Denied topics · Leave this out of the prompt

Four policy families on the gateway: identity is who is calling, field is what they may see, row is which records, and action is what they may do. A semantic check for PII and denied topics sits at the same edge.

Mermaid source

flowchart TD
  A["Identity: who is calling"] --> B["Field: which attributes"]
  B --> C["Row: which records"]
  C --> D["Action: read, write, or refuse"]
  D --> E["Semantic: PII and denied topics"]

Your agent surface

Every connector, MCP server, and approver, named.

Who asks

The question behind the brief.

Example
Risk officer at a DIFC wealth manager
Today: An advisor-assist agent is planned and DIFC Regulation 10 applies.
After: The agent is in a register with its tier, its owner and its off switch.
Example
CIO at a US health insurer
Today: Teams have connected agents to internal tools without a list.
After: Every MCP server and the scopes it holds are listed and cut to least privilege.
Example
Head of HR systems at a Riyadh group
Today: An agent will read HR data to answer staff questions.
After: Field and row policies on the gateway keep salary and medical fields out.

Examples of the work, not case studies. Roles and cities are typical of the desks we see.

Integrations

HR and ops into Salesforce, still governed.

People, payroll, and tickets already live somewhere. The gateway still sits on the hop.

Questions

What we actually say.

Is MCP a security model?: No. MCP is how an agent reaches a tool. Identity, least privilege, and audit still sit in your IdP, Salesforce sharing, and the gateway.
Who keeps the controls current after go-live?: We can, on the Run plan: set hours a month, a weekly read of transcripts and errors, and a written monthly note. The plans and prices are on the What you get page.

From the blog

Can you stop an agent in two hours?

UAE CIO data, what DIFC Regulation 10 adds, and a five-step stop drill.

The brief · one email

Tell us what is already running.

We will mark which agents have no policy family, and which should not exist. The assessment is free: a short read of the org and a written note. If you go ahead, the first build week is a fixed AED/SAR 14,900 (about USD 4,050), credited in full when you sign the build within 60 days. The honest no: if the work is a software factory with a named delivery date, we are the wrong partner.