@jsbarretto I'm sure there's plenty material on the topic, but from what I recall of fixed-function pipelines, clipping is usually done in NDC space, which means it happens after the view frustum transformation. The advantage is that clipping is much easier, because you only have to compare coordinates to [-1,1].

You'll still have to process all edges of all triangles, though:
1. Loop over all triangles
2. Test which vertices of the triangle are outside or inside the NDC cube.
3. If all are outside, cull the triangle. Skip to next.
4. If all are inside, no culling/clipping is needed. Skip to next.
5. If one is inside, calculate the intersection of the clipped edges with the NDC cube. This is easy, because you can fix one coordinate to the clip plane (x, y or z = -1 or 1) and calculate the others with the help of to the line equation (r = a + t • (b - a)). Generate two new vertices with the calculated coordinates and replace the triangle with a new one created from the new vertices. Skip to the next.
6. If two are inside, calculate the intersection vertices as in step 5, but replace the triangle with 2 new ones forming a trapezoid. Skip to the next.

You could keep track of the unused vertices and remove them at the end, but that will require additional processing time. They'll be ignored by the rasterizer if they're not used in any triangle, but if memory usage or bandwidth is a concern, you might still have to do that.