We present a formulation of the singles and doubles coupled-cluster (CCSD) method suitable for distributed memory parallel computer architectures. Results from test calculations using up to 128 nodes of the Intel i860 hypercube are presented. Although the new algorithm is highly parallel, the asymptotic speed is quickly reached due to the input/output (IO) demands on the concurrent file system. To alleviate these IO problems, as well as disk storage requirements, we propose a "semi-direct" CCSD method whereby the atomic orbital integrals are evaluated in each iteration of the procedure in order to determine the contribution from certain terms. Using CCSD(T)/TZ2P optimized geometries, the calculations show that the Born-Oppenheimer groundstate potential energy surface yields a linear structure for fulminic acid, HCNO, in agreement with the semirigid bender analysis by Bunker, Landsberg and Winnewisser.