As centralized social media platforms face growing concerns, more users are seeking greater control over their social feeds and turning to decentralized alternatives such as Mastodon. The decentralized nature of Mastodon creates unique opportunities for customizing feeds, yet user perceptions and curation strategies on these platforms remain unknown. This paper presents findings from a two-part interview study with 21 Mastodon users, exploring how they perceive, interact with, and manage their current feeds, and how we can better empower users to personalize their feeds on Mastodon. We use the qualitative findings of the first part of the study to guide the creation of Braids, a web-based prototype for feed curation. Results from the second part of our study, using Braids, highlighted opportunities and challenges for future research, particularly in using seamful design to enhance people's acceptance of algorithmic curation and nuanced trade-offs between machine learning-based and rule-based curation algorithms. To optimize user experience, we also discuss the tension between creating new apps and building add-ons in the decentralized social media realm.