Background: Many people diagnosed with schizophrenia, bipolar or other psychoses in England receive the majority of their healthcare from primary care. Primary care practitioners may not be well equipped to meet their needs and there is often poor communication with secondary care. Collaborative care is a promising alternative model but has not been trialled specifically with this service user group in England. Collaborative care for other mental health conditions has not been widely implemented despite evidence of its effectiveness. We carried out a formative evaluation of the PARTNERS model of collaborative care, with the aim of establishing barriers and facilitators to delivery, identifying implementation support requirements and testing the initial programme theory. Methods: The PARTNERS intervention was delivered on a small scale in three sites. Qualitative data was collected from primary and secondary care practitioners, service users and family carers, using semi-structured interviews, session recordings and tape-assisted recall. Deductive and inductive thematic analysis was carried out; themes were compared to the programme theory and used to inform an implementation support strategy. Results: Key components of the intervention that were not consistently delivered as intended were: interaction with primary care teams, the use of coaching, and supervision. Barriers and facilitators identified were related to service commitment, care partner skills, supervisor understanding and service user motivation. An implementation support strategy was developed, with researcher facilitation of communication and supervision and additional training for practitioners. Some components of the intervention were not experienced as intended; this appeared to reflect difficulties with operationalising the intervention. Analysis of data relating to the intended outcomes of the intervention indicated that the mechanisms proposed in the programme theory had operated as expected. Conclusions: Additional implementation support is likely to be required for the PARTNERS model to be delivered; the effectiveness of such support may be affected by practitioner and service user readiness to change. There is also a need to test the programme theory more fully. These issues will be addressed in the process evaluation of our full trial. Trial registration: ISRCTN95702682, 26 October 2017.