Tiktok Agent Publisher

What davidmosiah/tiktok-agent-publisher MCP server does

The davidmosiah/tiktok-agent-publisher MCP server connects agent workflows to TikTok's official Content Posting API. It exposes MCP tools and a command-line interface for checking connection readiness, reviewing privacy boundaries, creating video publish jobs, checking publish status, and listing recent videos.

The project is designed around an agent workflow rather than an interactive browser session. Responses use structured JSON so an agent can inspect results, retry operations, and record audit information. Readiness checks report whether client credentials, access tokens, refresh tokens, and media hosting prerequisites are present without returning secret values.

How it works

The package runs locally over stdio by default. An optional stateless Streamable HTTP mode listens on loopback and provides a health endpoint plus an MCP endpoint. The MCP surface includes tiktok_agent_manifest, tiktok_connection_status, tiktok_privacy_audit, tiktok_publish_video, tiktok_publish_status, and tiktok_list_videos.

A typical workflow starts with connection status and the privacy audit. In dry-run mode, publishing validates the job, caption, and metadata without calling the live API. It returns a synthetic publish identifier, and status polling reports a dry-run state. Listing videos also returns a valid empty result shape, allowing downstream workflow code to be tested before credentials are available.

Live operation requires OAuth-related configuration and an explicit change from the default dry-run setting. The project also supports an auth-url CLI command for building an OAuth URL from a redirect URI.

Setup and configuration

Install the npm package globally, or invoke it through npx. MCP clients can run the package with a stdio command. The README also documents a direct tiktok-agent-mcp entry point and an HTTP launch form using the package's MCP executable.

Configuration is read from environment variables or a local .env file. The documented TikTok values include TIKTOK_CLIENT_KEY, TIKTOK_CLIENT_SECRET, TIKTOK_ACCESS_TOKEN, and TIKTOK_REFRESH_TOKEN. TIKTOK_DRY_RUN controls the safety mode; it should remain enabled while validating an agent workflow and be set to false only when live calls are intended. HTTP deployments can set TIKTOK_MCP_HOST, TIKTOK_MCP_PORT, and TIKTOK_MCP_TRANSPORT.

Do not commit .env, token files, or the .agent-data/ directory. The readiness command indicates what is missing and whether live publishing or direct video upload is available.

Limitations and notes

The davidmosiah/tiktok-agent-publisher MCP server does not remove TikTok API or OAuth prerequisites. A dry-run can be exercised without credentials, but live publishing depends on valid TikTok configuration and an OAuth flow. The README specifically requires explicit confirmation before live direct posting and says that agents should publish only content the user owns or is permitted to publish.

Secrets are intended to stay local: tools do not return token values. HTTP mode is documented as loopback-only, so the provided setup is aimed at local agent runtimes rather than an openly hosted service. The material also describes media-hosting readiness, including a public media base URL and Supabase Storage status, but does not establish that every media workflow is available in every configuration.

The package is MIT-licensed and is presented as open-source tooling. No hosted service, paid plan, or paid API requirement is described in the provided material.