MODEL-INDEPENDENT / SUPERVISED API-WRITE GATEWAY

MOVE YOUR AGENT
FROM READ TO WRITE.

ActionDock is a human-in-the-loop approval gateway for AI-agent API writes: an MCP server and HTTP API that puts deterministic policy, owner approval of the exact request, and server-held credentials between an agent's tool call and a real API write. It binds the exact resolved target, query, and body into the approval preview, executes with a server-held credential, and returns an execution receipt.

Try the free sandbox Map your first safe write ↗Inspect the transaction boundary ↓

AGENT INTENTPATCH PROJECT RECORDANY AGENT · MCP OR HTTP

INTENT01

integration.execute

Intent captured

POLICY02

destination · method · path

Policy allowed

APPROVAL03

owner.approval

Request approved

EXECUTE04

provider.request

Provider responded

RECEIPT05

job + provider response

Receipt recorded

BOUNDARY 01 / FIRST SAFE WRITEIntent becomes an owner-approved request

Intent → deterministic policy → owner approval → execution receipt

SEE THE BOUNDARY

THE SUPERVISED WRITE CONTRACT

Intent enters.
Receipt returns.

ActionDock is the API-write boundary between an existing agent and a compatible public REST API. It is designed for low- to moderate-volume changes that deterministic policy and a person should clear before execution.

01 / INTENT

Accept one explicit change

The agent calls integration.execute with a connection, method, path, query, and body. It receives a job reference—not the credential behind the connection.

02 / POLICY

Evaluate deterministic rules

ActionDock enforces destination, method, path, network, and execution limits on the server. It then binds the connection version, resolved target, and exact query and body into the transaction preview. The model does not decide whether its own request is allowed.

03 / APPROVAL

Bind approval to this request

A signed-in workspace owner reviews the exact transaction preview. Execution stays bound to the request version the owner approved.

04 / EXECUTION

Broker the credential

Only ActionDock sends the approved request with the server-held bearer token or API key. The agent never receives the secret.

05 / RECEIPT

Record what ActionDock observed

The receipt includes execution state, timestamps, request metadata, provider response, and error details. If the provider may have accepted a request but its response was lost, the job becomes execution_unknown: blind retry stays blocked, and an operator must reconcile downstream state before deciding what happens next. The receipt is an execution record, not proof that the business outcome completed.

START WITH THE WRITE STILL BLOCKED BY HAND

Make one real action
safe to delegate.

Start with a specific POST, PATCH, PUT, or DELETE that an agent can prepare but cannot safely execute today. Keep every supporting read or analysis subordinate to that transaction.

01 / WRITEFirst safe API write

Put deterministic policy, exact-request approval, credential brokering, and an execution receipt around one real change.

READ ↗02 / CONTROLVisible execution state

Track the approval state, provider response, and errors for the proposed change.

READ ↗03 / PREPAREStructured input and checks

Extract fields and apply explicit rules before the agent proposes a write.

READ ↗04 / SUPPORTDocument evidence

Analyze supplied material when a contract, policy, or request informs the change.

READ ↗05 / SUPPORTResearch and handoff

Gather source links or create a report when the reviewer needs more context.

READ ↗

ONE WRITE PATH + SIX SUPPORTING ACTIONS

Lead with the write.
Support the decision.

integration.execute is the product’s primary operation. The other six MCP and HTTP actions prepare, check, or document the transaction; they are not the product boundary by themselves.

01integration.execute
Submit one bounded API write for owner review, then record the provider response.
CORE WRITE
02integration.fetch
Read from a permitted HTTPS method and path before or after the change.
SUPPORTING READ
03data.extract
Prepare caller-defined fields with supporting evidence.
SUPPORTING
04data.validate
Apply explicit rules and return machine-readable failures.
SUPPORTING
05document.analyze
Create a cited brief from supplied business text.
SUPPORTING
06research.web
Research a bounded question with source links.
SUPPORTING
07report.create
Persist a Markdown or JSON handoff artifact.
SUPPORTING

FIRST SAFE WRITE / LOCAL SIMULATION

Submit intent.
Inspect the boundary.

Run integration.execute, inspect the exact transaction preview, and simulate owner approval. This local demo never contacts a provider or changes external data.

REAL-PRODUCT FIT

Public REST APIs with bearer-token or API-key authentication; low- to moderate-volume writes that a person reviews.

ACTION RUNNERDEMO / LOCAL SIMULATION

Public sandbox NO KEY · NO NETWORK WRITE · $0

ActionInput JSON

Ready · no key required

{
  "mode": "simulation",
  "status": "ready",
  "action": "integration.execute",
  "message": "Preview the proposed write. Nothing leaves this page."
}

FIXED SAAS SUBSCRIPTIONS

Choose the
operating boundary.

Every plan includes the complete action catalog, API keys, approvals, workflow history, signed callbacks, and a monthly processing limit. There are no automatic overage charges and no purchasable usage balance. See the full pricing page ↗

01 / SUBSCRIBEOne fixed plan

Monthly or annual checkout.

02 / CONNECTApprove destinations

Methods, paths, and secrets stay server-side.

03 / EXECUTERun bounded work

All seven actions are included.

04 / CONTROLNo automatic overage

Work pauses at the monthly limit.

Launch

A first production agent workflow.

$49PER MONTH

3 API CONNECTIONS

Owner-approved destinations and credentials.

250 MB REPORTS

Full action catalog, processing limit, and email support.

Choose Launch ↗

Growth

Several internal systems and workflows.

$149PER MONTH

10 API CONNECTIONS

More approved systems under one workspace.

1 GB REPORTS

Higher processing limit and priority support.

Choose Growth ↗

Scale

Broad agent operations across a team.

$399PER MONTH

30 API CONNECTIONS

Wider system coverage with the same controls.

3 GB REPORTS

Highest processing limit and rollout review.

Choose Scale ↗

Enterprise

Custom security, support, and deployment scope.

ANNUALWRITTEN QUOTE

$12K–$120K+

Exact scope and price disclosed before purchase.

CONTRACTED TERMS

Security review, onboarding, and support scope. Download pricing guide ↗

What is ActionDock?

ActionDock is a human-in-the-loop approval gateway for AI-agent API writes: an MCP server and HTTP API that puts deterministic policy, owner approval of the exact request, and server-held credentials between an agent's tool call and a real API write. Every approved write returns an execution receipt, and a lost provider response becomes execution_unknown for operator reconciliation instead of a blind retry.

How do I connect an agent?

Official MCP server: https://actiondock.app/mcp (remote Streamable HTTP — nothing to install). ActionDock (actiondock.app) is unrelated to the actiondock and @actiondock/* packages on npm and to “Action Dock” boat-lift or loading-dock businesses.

claude mcp add --transport http actiondock https://actiondock.app/mcp --header "Authorization: Bearer <ACTIONDOCK_API_KEY>"

What happens at the monthly limit?

New jobs pause until the next subscription period or a plan upgrade. ActionDock never creates an automatic overage charge.

What is not included?

ActionDock is not a communications, marketing, social-posting, payment-processing, media-generation, or autonomous decision service. It relays only the requests you approve to systems you already control, and it never holds funds, card data, or a provider balance.

Can I cancel?

Yes. Manage the subscription in the customer portal. Paid access continues through the applicable billing period unless the portal states otherwise.

YOUR FIRST SAFE WRITE

Start with the write
still done by hand.

Map a blocked write Read the agent guide ↗

MOVE YOUR AGENTFROM READ TO WRITE.

Intent enters.Receipt returns.