Connect AI tools without confusing visibility with entitlement.
MCP is an interface, not a new license tier.
It gives a compatible client a structured way to call selected TaiGong operations. It does not unlock licensed modules, convert unsupported channels into supported ones, or authorize a security exercise. The TaiGong backend remains the source of truth.
This separation is essential for safe automation. Tool metadata helps a model decide what to request. The backend decides whether the request is authenticated, permitted, valid for the current version, and allowed by the organization’s license and policy.
Use four control layers.
| Layer | Responsibility | What it does not prove |
|---|---|---|
| MCP client | Chooses and formats a tool request. | That the proposed action is authorized. |
| MCP adapter | Maps the request to a TaiGong API operation. | That a licensed backend feature is available. |
| TaiGong backend | Applies authentication, validation, tenant, license, and product rules. | That the operator’s organizational approval exists. |
| Human operator | Owns written scope, review, approval, and consequences. | That model output is automatically correct. |
The 1.3.4 scope is intentionally specific.
- API adapter: the MCP component is an integration-only layer over product operations.
- Inherited license checks: a licensed backend capability stays licensed when requested through MCP.
- Campaign creation: the current MCP create-campaign flow is email-only.
- SMS and enterprise IM: the current MCP surface can expose related listing operations; this is not the same as creating and launching those channel campaigns through MCP.
- Unsupported channels: Slack is not a simulation-delivery channel, and Microsoft Teams delivery is unsupported in the current product.
Review the complete v1.3.4 support matrix →
Make high-impact actions reviewable.
- Use a least-privilege TaiGong account or API token.
- Keep tenant and license boundaries enforced by the backend.
- Require written authorization before generating or launching any campaign.
- Preview targets, content, delivery profile, schedule, and landing behavior.
- Keep audit evidence for the operator, client, requested tool, and backend result.
- Handle model-proposed input as untrusted data and reject ambiguous scope.
An MCP client can help an operator work with TaiGong. It does not turn TaiGong 1.3.4 into an autonomous campaign Agent, and it does not make a visible tool safe to call without review.
v1.3.5 remains future until released.
Broader MCP launch material, larger tool catalogs, and conversational Agent workflows belong to roadmap or future-release communication. This page describes the current public version, 1.3.4. It will be revised when a later release is actually delivered and verified.