We experimentally study the integration of refugees into a national public works-based safety net program in Ethiopia, a policy approach increasingly viewed as a sustainable alternative to parallel, aid-financed assistance in contexts of protracted displacement. Leveraging an at-scale nationwide randomized controlled trial among 25,000 eligible refugee and host, across Ethiopia's four main refugee-hosting regions, we estimate the causal effects of integrated social protection -- comprised of remunerated public works (up to 25h per week and to 20% of baseline household expenditure) and livelihoods trainings -- on economic outcomes, well-being, and social cohesion. After 16 months, the program generates large gains in household income, assets, consumption, and labor supply, with limited crowding out of other activities for both refugees and hosts. It also leads to substantial improvements in well-being, including physical and mental health, increases in women’s empowerment, and improvements in child nutrition. Social cohesion improves meaningfully, with effects concentrated among host households. To study mechanisms underlying social cohesion impacts, we examine the effects of randomly assigning refugees and hosts to mixed public work groups and after-work socializing through teas and dinners. Mixed team members reallocate social and economic interactions toward out-group members, and mixing increases pro-refugee attitudes among hosts, without significant changes in team productivity. Leveraging the random allocation of neighborhoods into higher vs. lower share of mixed groups, we document that social cohesion impacts of out-group exposure spill over to non-exposed neighbors, suggesting the program improved community relations more broadly.