Summary
Posting from the client you are in, with a tool list short enough to audit in one look.
Two tools, and one of them publishes to a live account: there is no draft, no edit and no delete in the surface, so whatever the agent decides to post is what ships. The read side is a query with a result count, which is enough to check what a thread is saying before you reply to it and not enough to do timeline analysis.
What it is
A two-call Twitter client: one write that posts a tweet or replies to an existing tweet id, and one read that searches tweets for a query.
What you get
- Posting a tweet from wherever the model is already working
- Replying to a specific tweet by its id, with the reply text in the same call
- A search over tweets for a query, returning the number of results you ask for
Requirements
Four Twitter API credentials: API_KEY, API_SECRET_KEY, ACCESS_TOKEN and ACCESS_TOKEN_SECRET.
Setup effort
One command plus a key — npx -y @enescinar/twitter-mcp, then supply credentials