Background Digital health interventions have the potential to improve the delivery of psychoeducation to people with mental health problems and their relatives. Despite substantial investment in the development of digital health interventions, successful implementation into routine clinical practice is rare. Objectives Use the implementation of the Relatives’ Education And Coping Toolkit (REACT) for psychosis/bipolar disorder to identify critical factors affecting uptake and use, and develop an implementation plan to support the delivery of REACT. Design This was an implementation study using a mixed-methods, theory-driven, multiple case study approach. A study-specific implementation theory for REACT based on normalisation process theory was developed and tested, and iterations of an implementation plan to address the key factors affecting implementation were developed. Setting Early-intervention teams in six NHS mental health trusts in England (three in the north and three in the south). Participants In total, 281 staff accounts and 159 relatives’ accounts were created, 129 staff and 23 relatives took part in qualitative interviews about their experiences, and 132 relatives provided demographic data, 56 provided baseline data, 21 provided data at 12 weeks’ follow-up and 20 provided data at 24 weeks’ follow-up. Interventions REACT is an online supported self-management toolkit, offering 12 evidence-based psychoeducation modules and support via a forum, and a confidential direct messaging service for relatives of people with psychosis or bipolar disorder. The implementation intervention was developed with staff and iteratively adapted to address identified barriers. Adaptations included modifications to the toolkit and how it was delivered by teams. Main outcome measures The main outcome was factors affecting implementation of REACT, assessed primarily through in-depth interviews with staff and relatives. We also assessed quantitative measures of delivery (staff accounts and relatives’ invitations), use of REACT (relatives’ logins and time spent on the website) and the impact of REACT [relatives’ distress (General Health Questionnaire-28), and carer well-being and support (Carer Well-being and Support Scale questionnaire)]. Results Staff and relatives were generally positive about the content of REACT, seeing it as a valuable resource that could help services improve support and meet clinical targets, but only within a comprehensive service that included face-to-face support, and with some additional content. Barriers to implementation included high staff caseloads and difficulties with prioritising supporting relatives; technical difficulties of using REACT; poor interoperability with trust information technology systems and care pathways; lack of access to mobile technology and information technology training; restricted forum populations leading to low levels of use; staff fears of managing risk, online trolling, or replacement by technology; and uncertainty around REACT’s long-term availability. There was no evidence that REACT would reduce staff time supporting relatives (which was already very low), and might increase it by facilitating communication. In all, 281 staff accounts were created, but only 57 staff sent relatives invitations. In total, 355 relatives’ invitations were sent to 310 unique relatives, leading to the creation of 159 relatives’ accounts. The mean number of logins for relatives was 3.78 (standard deviation 4.43), but with wide variation from 0 to 31 (median 2, interquartile range 1–8). The mean total time spent on the website was 40.6 minutes (standard deviation 54.54 minutes), with a range of 0–298 minutes (median 20.1 minutes, interquartile range 4.9–57.5 minutes). There was a pattern of declining mean scores for distress, social dysfunction, depression, anxiety and insomnia, and increases in relatives’ well-being and eHealth literacy, but no changes were statistically significant. Conclusions Digital health interventions, such as REACT, should be iteratively developed, evaluated, adapted and implemented, with staff and service user input, as part of a long-term strategy to develop integrated technology-enabled services. Implementation strategies must instil a sense of ownership for staff and ensure that they have adequate training, risk protocols and resources to deliver the technology. Cost-effectiveness and impact on workload and inequalities in accessing health care need further testing, along with the generalisability of our findings to other digital health interventions. Limitations REACT was offered by the same team running the IMPlementation of A Relatives’ Toolkit (IMPART) study, and was perceived by staff and relatives as a time-limited research study rather than ongoing clinical service, which affected engagement. Access to observational data was limited. Trial registration Current Controlled Trials ISRCTN16267685. Funding This project was funded by the National Institute for Health Research (NIHR) Health Services and Delivery Research programme and will be published in full in Health Services and Delivery Research; Vol. 8, No. 37. See the NIHR Journals Library website for further project information.