Let an agent post to X, Threads and LinkedIn — and prove what it did.

Issue a scoped token, expose a fixed operation catalogue over remote MCP or plain HTTP, and read a durable receipt for every attempt. You keep the accounts and the approval.

Connect an agent Read the agent guide Open workspace preview

root@poststeward:~$ why

PostSteward exists so a developer can stay inside the build while an agent keeps the product story, social engagement and GTM pipeline moving. The agent can work from project and GTM context it already has; PostSteward gives it the controlled publishing lane.

[SHOWCASE MODE] This public surface teaches the agent contract and mirrors the real workspace UI. It does not accept provider credentials, issue live tokens, create payments or execute social posts. Endpoint examples use <POSTSTEWARD_SERVICE_ORIGIN> until public effectful access is opened.

Three agent paths, one operation catalogue

Remote MCP, CLI shell calls and HTTP share one catalogue, one set of execution rules and the same receipts. WebMCP exposes the same scoped operations inside a supported browser workspace.

┌─ Remote MCP

{
  "mcpServers": {
    "poststeward": {
      "url": "<POSTSTEWARD_SERVICE_ORIGIN>/mcp",
      "headers": {
        "Authorization": "Bearer <agent token>"
      }
    }
  }
}

┌─ CLI via HTTP

curl -X POST \
  <POSTSTEWARD_SERVICE_ORIGIN>/api/operations/workspace_status \
  -H "Authorization: Bearer <agent token>" \
  -H "Content-Type: application/json" \
  -d '{}'

┌─ Browser WebMCP

// On a real connected workspace
const tools = await document.modelContext.getTools({
  fromOrigins: []
});

// Same names and scopes as MCP / HTTP

Rules the agent can rely on

Scoped tokens, not credentials

An agent receives a Bearer token limited to named operations with an expiry. Provider credentials stay in the workspace.

Exact copy, no rewriting

A campaign is immutable once created. What the agent submits is what gets published, to a verified account.

Declared consequences

Effectful operations require an idempotency key and state their effect before running. An uncertain publication is never retried blindly.

Receipts an agent can read

Every attempt returns a durable receipt with delivery status, verified identity and the published link when the provider confirms one.

Free · hosted

Direct the work

—Connect X, Threads and LinkedIn accounts with verified identity routing.
—Publish now or set explicit schedules, then cancel or replace them.
—Inspect durable receipts, capture available metrics and export everything.

No service subscription

Advanced · $5 per month

Keep the work moving

—Monitor reviewed product sources and replenish campaign inventory.
—Allocate future posts with spacing controls.
—Collect metrics on a schedule instead of on demand.

USD 5 per workspace each month

Built to be found and understood

Agents can use remote MCP, plain HTTP operations, CLI shell calls, or browser WebMCP in a supported browser. Every interface shares one operation catalogue and the same execution rules, including declared consequences and no blind retry after an uncertain publication.

Agent guide Machine-readable help OpenAPI Operation reference llms.txt agents.txt mcp.json

Provider API charges are separate. X requires a customer-funded developer application. Public purchases and effectful agent access remain closed on this showcase until launch acceptance is complete.