Background: Cancer early detection aims at reducing cancer deaths. Unfortunately, many established cancer screening technologies are not suitable for use in low- and middle-income countries (LMICs) due to cost, complexity, and dependency on extensive medical infrastructure. Methods: Nearly 10,000 participants [1959 cancer cases (containing more than nine common cancer types) and 7423 non-cancer cases] were divided into one training and two independent validation cohorts. One tube of peripheral blood samples was collected from each participant and quantified using a panel of seven selected protein tumour markers (PTMs) by a common clinical electrochemiluminescence immunoassay analyser. An algorithm named OncoSeek was established using artificial intelligence (AI) to distinguish cancer from non-cancer cases by calculating the probability of cancer (POC) index based on the quantification results of the seven PTMs and clinical information including sex and age of the subjects, and to predict the possible affected tissue of origin (TOO) for those who have been detected with cancer signal. Results: The conventional clinical method relied only on a single threshold for each PTM which would make a big problem when combining the results of those markers, the false positive rate would accumulate as the number of markers increased. Nevertheless, OncoSeek was empowered by AI technology to significantly reduce the false positive rate, increasing the specificity from 56·9% [95% confidence interval (CI): 55·8% to 58·0%] to 92·9% (95% CI: 92·3% to 93·5%). In all cancer types, the overall sensitivity of the OncoSeek test was 51·7% (95% CI: 49·4% to 53·9%). And the performance was generally consistent in the training and the two validation cohorts. The sensitivities ranged from 37·1% to 77·6% for the detection of the nine common cancer types (breast, colorectum, liver, lung, lymphoma, oesophagus, ovary, pancreas, and stomach), which account for ~59·2% of global cancer deaths annually. Furthermore, it has shown excellent sensitivity in several high-mortality cancer types for which there are lacking routine screening tests in the clinic, such as the sensitivity of pancreatic cancer was 77·6% (95% CI: 69·3% to 84·6%). The overall accuracy of TOO prediction in the true positives was 66·8%, which could assist the clinical diagnostic workup. Conclusions: OncoSeek significantly outperforms the conventional clinical method, representing a novel blood-based test for multi-cancer early detection (MCED) which is a non-invasive, easy, efficient, and robust approach. Moreover, the accuracy of TOO facilitates the follow-up diagnostic workup and this method is affordable (less than $25) and accessible requiring nothing more than a blood draw at the screening sites, which makes it acceptable and sustainable in LMICs.