The sandbox email detector, built to hand to an agent
Today silobase runs as a web scan: upload a Salesforce metadata export, get back a blast-radius manifest of every path that would send email. An MCP server that exposes the same detector to Claude, Cursor, and other agents is on the roadmap — not yet published.
There is no live silobase MCP endpoint today. When the server ships it will run the same deterministic detector that already powers the web scan — until then, the scan is the way to run it. We would rather leave this blank than point you at something that is not live.
- Server endpoint
- {{TODO: MCP server not yet published}}
- Registry listing
- {{TODO: MCP server not yet published}}
Upload a metadata export. Get the manifest back.
silobase is an upload-only static analyzer. It parses a Salesforce metadata export — a zip or SFDX source — and detects every email-sending path across Flows, workflow rules, email alerts, Apex senders, approval emails, and the org-wide sender and deliverability settings. From that pass it produces a blast-radius manifest, a safe-test plan, a deliverability checklist, and an exportable evidence ledger.
An agent gets the same value the same way — through the scan — while the MCP interface is still on the roadmap.
What an agent actually needs is a finding it can trust
The classification is deterministic. The language model only explains what the detector already found — so the output is path-backed, not guessed.
The detector owns risk classification, not a model. The same metadata export yields the same blast-radius manifest every run — an agent can rely on the finding instead of re-deriving it.
Each detected email path names its component — the exact Flow, workflow rule, email alert, Apex sender, or approval step. An agent gets an address to act on, not a summary.
The deterministic detector produces the findings. The Claude API is handed those findings — not your raw org data — and writes the plain-language explanation and drafts the safe-test plan.
Flows, workflow rules, email alerts, Apex senders, approval emails, plus org-wide sender and deliverability settings — mapped from one metadata export into one manifest.
What the planned MCP server would expose
The shape below is planned, not published. It describes the interface we intend to ship, backed by the detector that already runs in the web scan.
- A scan tool that takes a Salesforce metadata export and returns the blast-radius manifest — the same deterministic detector that runs behind the web scan.
- Structured rows an agent can read directly: each email path, its source component, its sender, whether it would send, and its evidence path.
- The same hard boundaries as the scan — metadata only, no sends, no writes back, upload-only heuristics.
- Endpoint & registry{{TODO: MCP server not yet published}}
What silobase does not do
An MCP interface would not widen the blast radius. Every boundary that governs the scan governs an agent calling it.
It reads a metadata export. It never reads email content, CRM records, or contact data.
It maps the paths that would send email. It never triggers one, and it never writes back to your org.
Recipient resolution is a heuristic — v1 is upload-only, with no live-org OAuth. It is not a live-org connection.
Not a Salesforce product, not endorsed by Salesforce, and not Mailtrap. It is a separate static analyzer.
Building an agent that needs this and want word when the MCP server ships? Tell us at support@silobase.io.
Run the detector today through the scan.
The free scan surfaces 1–2 risky paths with no signup. The $99 one-time report covers one org for one release cycle; the $39/mo release ledger tracks unlimited orgs release over release.