Achieving net zero CO2 emissions requires gigatonne-scale atmospheric CO2 removal (CDR) to balance residual emissions that are extremely difficult to eliminate. Marine CDR (mCDR) methods are seen increasingly as potentially important additions to a global portfolio of climate policy actions. The most widely considered mCDR methods are coastal blue carbon and seaweed farming that primarily depend on biological manipulations; ocean iron fertilisation, ocean alkalinity enhancement, and direct ocean capture that depend on chemical manipulations; and artificial upwelling that depends on physical manipulation of the ocean system. It is currently highly uncertain which, if any, of these approaches might be implemented at sufficient scale to make a meaningful contribution to net zero. Here, we derive a framework based on additionality, predictability, and governability to assess implementation challenges for these mCDR methods. We argue that additionality, the net increase of CO2 sequestration due to mCDR relative to the baseline state, will be harder to determine for those mCDR methods with relatively large inherent complexity, and therefore higher potential for unpredictable impacts, both climatic and non-climatic. Predictability is inherently lower for mCDR methods that depend on biology than for methods relying on chemical or physical manipulations. Furthermore, predictability is lower for methods that require manipulation of multiple components of the ocean system. The predictability of an mCDR method also affects its governability, as highly complex mCDR methods with uncertain outcomes and greater likelihood of unintended consequences will require more monitoring and regulation, both for risk management and verified carbon accounting. We argue that systematic assessment of additionality, predictability, and governability of mCDR approaches increases their chances of leading to a net climatic benefit and informs political decision-making around their potential implementation.