Recipes: hand these jobs to an agent

Each recipe lists the MCP tools in the order to call them. Paste one into your agent with the goal on top, or keep it as a reference. Connect the agent first with the MCP reference.

Triage the inbox every morning

1. list_conversations with status: "open".
2. get_conversation for each, to read the thread and see the person's page and browser.
3. For a question a published article answers, find it with list_help_articles and a query, then reply with the answer and the article's link.
4. For the rest, assign_conversation to whoever owns that area, and leave the reply to them.
5. Report back: what it answered, what it left, and why.

Let the agent answer only from published articles. If it can't point to one, a person should reply.

Ship a request and tell everyone who asked

1. list_posts with sort: "top" to find the request.
2. create_changelog_entry with a title, the body, and the request's id in post_ids.
3. Show the draft to a person.
4. publish_changelog_entry. teem marks the request shipped and emails everyone who voted for it, once.

Publishing is the step that sends email, so keep a person between steps 2 and 4.

Publish release notes from your deploy script

A deploy script is code, so this one uses REST. Store the key as a secret in your CI, never in the repository.

id=$(curl -s https://teem.so/api/v1/changelog \
  -H "Authorization: Bearer $TEEM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"changelog_entry":{"title":"Faster search","body":"Search results now load in under a second."}}' \
  | jq -r .data.id)

curl -s -X POST "https://teem.so/api/v1/changelog/$id/publish" \
  -H "Authorization: Bearer $TEEM_API_KEY"

Leave out the second call to keep the entry as a draft for someone to read first.

When a call fails, My agent gets an error from teem explains the error codes.