TikTok MCP
Give AI agents governed, auditable control over TikTok campaigns and creative
A Model Context Protocol server wraps TikTok's Business, Marketing, and Reporting APIs in typed tools an autonomous agent can call safely. Through it, agents can read performance, adjust budgets and bids, launch or pause ads, and pull creative insights within guardrails we define — so optimization runs continuously with a human approving the moves that matter.
Query spend, ROAS, CPA, watch-time, and creative-level metrics across campaigns, ad groups, and ads via the Reporting API.
Create, pause, resume, and rebudget campaigns and ad groups, and shift bids or targeting within approved bounds.
Upload assets, register Spark Ads via authorization codes, and rotate or retire fatiguing creative variants.
Update custom and exclusion audiences and manage product catalog entries for Shopping campaigns.
An agent watches ROAS and pacing hourly, reallocating budget from fatiguing ad groups to winners inside caps we set, and escalating any move above a threshold for human sign-off.
When frequency climbs and watch-time drops, the agent flags the ad, drafts fresh hook and caption variants from winning attributes, and queues them for approval before launch.
An agent pulls TikTok metrics alongside other paid channels each morning, reconciles them against CRM revenue in our warehouse, and posts a narrated performance brief to the team.
Only within limits you approve. We scope the MCP tools to specific accounts, cap budget and bid changes, and route anything beyond the threshold to a human approval step with a full audit log of every action.
Every write goes through typed tools with validation, rate limits, and logging. Changes are recorded so they can be reviewed and rolled back, and read-only workflows run freely while write actions stay gated.
No. It removes the repetitive monitoring and pacing work so your team focuses on strategy and creative. The agent proposes and executes routine moves; people set the guardrails and own the big decisions.
