Why Ducto — vs official platform MCPs and building your own

Why Ducto

The access already exists. The control doesn't.

Official platform MCPs and DIY integrations both answer “can my agent reach the account?” Ducto answers the question that actually blocks rollout: “what stops it from doing the wrong thing?”

Connect your first platform Read the security architecture

Three ways to wire an agent

Official MCPs, build it yourself, or Ducto

“Official platform MCPs” means the read/write servers platforms ship themselves. “Build it yourself” means a gateway your team writes and operates.

What you need	Official platform MCPs	Build it yourself	Ducto
Read the account in conversation
Write access to campaigns and budgets	Direct and unlimited	You implement it	Governed, per policy
Numeric caps on budget and bid writes	—	You build and maintain	From the policy row, server-side
Preview before apply	—	You build and maintain	Fingerprint-matched, enforced
Human approval on flagged classes	—	You build and maintain	Approval inbox, selective
Audit trail of every call	Platform change history only	You build and maintain	Hash-chained, append-only, exportable
Token scoping per platform and tool class	—	You build and maintain	Built in, revocable
Platforms behind one endpoint	One server per platform	You integrate each	16 platforms, 226 tools, one URL
OAuth apps, token refresh, rate limits	Per-platform setup	You operate it	Hosted — connect and go

Official servers are good software — for access. They are not designed to be a control layer, and platform change history is not an agent audit trail.

The short version

Three sentences, no adjectives

A prompt is not a policy — anything a model reads, a hijacked model can ignore. Ducto enforces after the model, before the platform.

Caps, previews, approvals, and the audit chain live server-side in Postgres. There is no code path from tool input to a platform write that bypasses them.

Building this yourself is a real project: 15 OAuth apps, a token vault, a policy engine, an approval queue, an append-only ledger — and then maintaining all of it.

FAQ

Fair pushback

Aren't the official MCPs enough for read-only work?

For pure reporting on one platform, they can be. The moment you want writes — or one endpoint across platforms, or an answer for the security review — you need the control layer. Ducto is also read-only by default, so starting there costs nothing.

We could build this in a sprint.

You could build a proxy in a sprint. Caps that survive a hostile prompt, previews matched by fingerprint, a selective approval queue, and a hash-chained ledger that even the database owner can't edit — that's the part that takes the year, and the part auditors ask about.

Does Ducto slow the agent down?

Reads and within-cap writes execute immediately. Only the tool classes you flag wait for a human — approval is selective by design.

What if we outgrow it?

Your policies are rows, your audit log is exportable, and tokens are revocable at any time. Leaving is a config change, not a migration project — which is exactly why it's safe to start.

Grant the access you've been sitting on

Read-only in minutes. Writes when you've set your caps. See all 16 platforms.

Get started free See pricing