Overview
Custom AI Agent Builder is part of the Postly AI Agents system. It helps teams move from manual content work to governed, repeatable marketing workflows that can analyze, recommend, generate, schedule, and report.
Allow users to create specialized agents using goals, instructions, sources, tools, permissions, triggers, and review rules.
Agent Setup Flow
- Choose agent: Select the agent that matches the workflow or business goal.
- Configure sources: Choose the platforms, feeds, products, blogs, analytics, or other data sources the agent can use.
- Connect MCP/API if needed: Authorize workspace access for external AI clients or developer workflows.
- Choose permissions: Control read, write, scheduling, publishing, and admin access.
- Set trigger: Run manually, on a schedule, or when an event is detected.
- Set review rules: Require approval, review first runs, or allow auto-publishing when appropriate.
- Test and activate: Run a test, inspect the output, then activate the agent.
Recommended default
Inputs and Data Sources
- Agent name
- Goal
- Instructions
- Restrictions
- Sources
- Tools
- Triggers
Agent Actions
- Run workflows
- Create drafts
- Schedule content
- Generate reports
- Send outputs for approval
Outputs
- Custom workflows
- Drafts
- Reports
- Run history
- Activity logs
Permissions
| Permission Type | Examples |
|---|---|
| Read | Read analytics, posts, scheduled posts, social accounts, media, blogs, RSS feeds, ecommerce products, inventory, discounts, and brand kits. |
| Write | Create drafts, update drafts, generate content, generate images, schedule content, and create campaigns. |
| Publish | Publish content and approve content when explicitly allowed. |
| Admin | Manage agents, integrations, MCP connections, and API access. |
Governance and Safety
- Scoped permissions
- Approval modes
- Run history
- Admin controls
Agent workflows should create clear run history with started time, completed time, status, sources used, actions taken, drafts created, posts scheduled, approvals pending, and errors.
MCP and API Access
AI Agents can be used inside Postly, through MCP clients such as ChatGPT, Claude, Cursor, Windsurf, and custom MCP clients, or through API-key based workflows. All connections should use scoped permissions and activity logs.
Agents should not own raw third-party API credentials. Postly should manage OAuth, API credentials, token refresh, rate limits, webhooks, API versioning, error handling, and credential storage.
Metrics to Watch
- Activation rate: How many users configure and activate an agent.
- First successful run: Whether the agent creates a useful report, draft, campaign, or recommendation.
- Approval rate: How many agent outputs are approved instead of rejected.
- Editing rate: How much users change agent-generated output before publishing.
- Scheduled or published output: Whether agent work turns into actual content operations.
Editor rule
FAQ
- Can this agent publish automatically?
- Yes, only when the workspace explicitly enables Auto Publish and the agent has publish permissions.
- Can this agent connect to external APIs directly?
- No. Agents consume Postly tools. Postly integrations manage OAuth, tokens, rate limits, webhooks, and credential storage.
- Can teams require approval?
- Yes. Approval workflows should be available for all agents and recommended for high-risk workflows.
Next Steps
- Choose the agent and connect the required sources.
- Set permissions, triggers, and review rules.
- Run a test and inspect the output.
- Activate the agent and monitor the Activity Center.