
LamaTok
io.github.subzeroidv1.1.2Updated Oct 6, 2026
TikTok data via LamaTok: profiles, videos, comments, followers, hashtags, search, downloads.
Overview
Lets an assistant pull TikTok data — profiles, videos, comments, followers, hashtags, search results and download links — through the LamaTok API.
- What it does
- Generates MCP tools at startup from the LamaTok OpenAPI spec, mapping each tool to a REST endpoint such as get_v1_user_by_username or get_v1_media_comments_by_id. The default core set exposes about 23 curated tools covering profiles and audience, videos and comments, download links, and hashtags and search; setting LAMATOK_TOOLS=all exposes every non-deprecated endpoint. Tool arguments map to the endpoint's query and path parameters, and responses are returned as raw JSON text.
- When to use it
- Use it when an assistant needs TikTok profile, video, comment, follower, hashtag or search data, or download links, for research, content monitoring or reporting. It is a read-only data source rather than a publishing or account-management tool.
- Requirements
- Runs locally over stdio, typically via npx lamatok-mcp, so Node.js is needed. A LamaTok access key is required in the LAMATOK_KEY environment variable, sent as the x-access-key header; keys come from lamatok.com. Network access to api.lamatok.com is needed, and the server fetches the OpenAPI spec once at startup. Optional variables configure base URL, tool set, tag filters, timeouts and response size limits.
Installation
In SourceWeft
- Open LamaTok in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
lamatok-mcp
[npm version] [npm downloads] [License: MIT]
MCP server for LamaTok — TikTok data API. Available on npm: lamatok-mcp.
Generates MCP tools from the LamaTok OpenAPI spec at startup. Tools map 1:1 to REST endpoints (GET /v1/user/by/username → get_v1_user_by_username). By default you get a core set of ~23 tools, one per task, each with a description that tells the assistant when to use it; LAMATOK_TOOLS=all exposes every non-deprecated endpoint.
Get 100 Free API Requests
Sign up with this link and get 100 free LamaTok requests — no credit card required. Enough to wire up the MCP server, try a few prompts in Claude/Cursor/Codex, and evaluate the data quality before committing.
Quick start
- Get an API key at lamatok.com.
- Add the server to your AI assistant.
- Ask your assistant something like:
- "Get the TikTok profile for @nasa."
- "List the last 10 videos by user_id 6707206320333226502."
- "Find recent TikTok videos for the hashtag
photography."
Claude Code
Claude Desktop
Add to claude_desktop_config.json:
Cursor / Windsurf
Same shape as Claude Desktop — put the block under mcpServers in the app's MCP config file.
Zed
Add to ~/.config/zed/settings.json:
OpenAI Codex
Append to ~/.codex/config.toml:
Tools
Tools are generated at startup from the live LamaTok OpenAPI spec. The default core set keeps one endpoint per task and drops version duplicates:
Core tools carry hand-written descriptions (what the tool does, when to prefer a sibling, pagination, billing) and every tool is annotated read-only. The list lives in src/curated.ts.
Set LAMATOK_TOOLS=all to expose every non-deprecated endpoint instead — same tool names as before, so existing prompts keep working. Tool names mirror their endpoint (GET /v1/user/by/username → get_v1_user_by_username); call tools/list over MCP for the current list with parameter schemas. /sys, Legacy, and System tag groups are excluded in both modes.
Configuration
Legacy, System, and /sys tags are excluded by default. Deprecated operations are also skipped.
If LAMATOK_URL points to a host other than api.lamatok.com, the server prints a warning on startup — your key will be sent there, so only use it for a self-hosted or proxied LamaTok.
Requests are sent with User-Agent: lamatok-mcp/<version>.
How it works
Tool arguments map to the endpoint's query and path parameters. The response body is returned as-is (JSON text). Non-2xx responses are surfaced as tool errors with the HTTP status and body.
Development
Run in watch mode:
Run tests (unit + stdio smoke tests against a local mock server, no network/API key required):
License
MIT
Source: README.md at commit d1e33e9
Tools
0Version history
1- v1.1.2LatestOct 6, 2026

