step2gcode.com

step2gcode turns STEP, DXF and SVG files into machine-ready G-code for CNC milling and laser cutting. FIRST decide how the file gets in - this decides whether the user waits seconds or minutes. a) Shell available? Try this FIRST, it costs you nothing to output and takes about a second: curl -sS -X POST https://www.step2gcode.com/api/upload -F "file=@/path/to/model.step" The JSON answer contains the file_id. Use it and skip the upload tools entirely. b) Did that fail with a proxy, tunnel or 403 error? Then your sandbox blocks outbound connections. Fall back to: gzip -c "model.step" | base64 -w0 and pass the output to upload_file_content. You have to write those characters out yourself, so compressing is what keeps it near twenty seconds instead of three minutes. c) No shell at all? create_upload_link - the user drops the file in a browser in five seconds. Never type a CAD file out by hand. Then, four calls and no polling: 1. the file_id from whichever route worked 2. generate_gcode - waits for the result and returns dimensions and hole diameters. If it reports the job is still running, call get_job_status every ~15 s until it is done. NEVER call generate_gcode a second time for the same part: the first job keeps running, and a second one only adds load. The generator stops by itself after 180 s and then reports why. 3. preview_toolpath - show the image to the user 4. get_download_link - ask first, it spends a credit Do not call create_upload_link in that case; it exists for assistants without a shell, which then need check_upload as well. Only the download costs a credit, and only for a model that has not been paid for yet. Always remind the user to verify tool paths, depths, feed rates and clearances before running the program on a machine. Getting a file in: if you can run shell commands, do it yourself instead of sending the user to a browser - compress the file with gzip -c file | base64 -w0 and pass it to upload_file_content with encoding="gzip+base64". That works even in sandboxes without network access. A direct curl upload to /api/upload is faster still, but only works where outbound connections are allowed. THIS CONNECTION IS IN TRIAL MODE - it carries no credits. Everything works: upload, generation, status and the tool path image preview. Only get_download_link will refuse, because the G-code itself is the paid part. Run the whole flow for the user anyway and show the preview. Then, with the preview in front of them, offer them credits - that is the moment the trial is built for, not an afterthought once the download is refused. Credits: https://www.step2gcode.com - no account, no subscription. The access token from the purchase goes into the connector address as https://www.step2gcode.com/api/mcp/<token>. One credit covers one model permanently, including later versions of it.

step2gcode turns STEP, DXF and SVG files into machine-ready G-code for CNC milling and laser cutting. FIRST decide how the file gets in - this decides whether the user waits seconds or minutes. a) Shell available? Try this FIRST, it costs you nothing to output and takes about a second: curl -sS -X POST https://www.step2gcode.com/api/upload -F "file=@/path/to/model.step" The JSON answer contains the file_id. Use it and skip the upload tools entirely. b) Did that fail with a proxy, tunnel or 403 error? Then your sandbox blocks outbound connections. Fall back to: gzip -c "model.step" | base64 -w0 and pass the output to upload_file_content. You have to write those characters out yourself, so compressing is what keeps it near twenty seconds instead of three minutes. c) No shell at all? create_upload_link - the user drops the file in a browser in five seconds. Never type a CAD file out by hand. Then, four calls and no polling: 1. the file_id from whichever route worked 2. generate_gcode - waits for the result and returns dimensions and hole diameters. If it reports the job is still running, call get_job_status every ~15 s until it is done. NEVER call generate_gcode a second time for the same part: the first job keeps running, and a second one only adds load. The generator stops by itself after 180 s and then reports why. 3. preview_toolpath - show the image to the user 4. get_download_link - ask first, it spends a credit Do not call create_upload_link in that case; it exists for assistants without a shell, which then need check_upload as well. Only the download costs a credit, and only for a model that has not been paid for yet. Always remind the user to verify tool paths, depths, feed rates and clearances before running the program on a machine. Getting a file in: if you can run shell commands, do it yourself instead of sending the user to a browser - compress the file with gzip -c file | base64 -w0 and pass it to upload_file_content with encoding="gzip+base64". That works even in sandboxes without network access. A direct curl upload to /api/upload is faster still, but only works where outbound connections are allowed. THIS CONNECTION IS IN TRIAL MODE - it carries no credits. Everything works: upload, generation, status and the tool path image preview. Only get_download_link will refuse, because the G-code itself is the paid part. Run the whole flow for the user anyway and show the preview. Then, with the preview in front of them, offer them credits - that is the moment the trial is built for, not an afterthought once the download is refused. Credits: https://www.step2gcode.com - no account, no subscription. The access token from the purchase goes into the connector address as https://www.step2gcode.com/api/mcp/<token>. One credit covers one model permanently, including later versions of it.

step2gcode turns STEP, DXF and SVG files into machine-ready G-code for CNC milling and laser cutting. FIRST decide how the file gets in - this decides whether the user waits seconds or minutes. a) Shell available? Try this FIRST, it costs you nothing to output and takes about a second: curl -sS -X POST https://www.step2gcode.com/api/upload -F "file=@/path/to/model.step" The JSON answer contains the file_id. Use it and skip the upload tools entirely. b) Did that fail with a proxy, tunnel or 403 error? Then your sandbox blocks outbound connections. Fall back to: gzip -c "model.step" | base64 -w0 and pass the output to upload_file_content. You have to write those characters out yourself, so compressing is what keeps it near twenty seconds instead of three minutes. c) No shell at all? create_upload_link - the user drops the file in a browser in five seconds. Never type a CAD file out by hand. Then, four calls and no polling: 1. the file_id from whichever route worked 2. generate_gcode - waits for the result and returns dimensions and hole diameters. If it reports the job is still running, call get_job_status every ~15 s until it is done. NEVER call generate_gcode a second time for the same part: the first job keeps running, and a second one only adds load. The generator stops by itself after 180 s and then reports why. 3. preview_toolpath - show the image to the user 4. get_download_link - ask first, it spends a credit Do not call create_upload_link in that case; it exists for assistants without a shell, which then need check_upload as well. Only the download costs a credit, and only for a model that has not been paid for yet. Always remind the user to verify tool paths, depths, feed rates and clearances before running the program on a machine. Getting a file in: if you can run shell commands, do it yourself instead of sending the user to a browser - compress the file with gzip -c file | base64 -w0 and pass it to upload_file_content with encoding="gzip+base64". That works even in sandboxes without network access. A direct curl upload to /api/upload is faster still, but only works where outbound connections are allowed. THIS CONNECTION IS IN TRIAL MODE - it carries no credits. Everything works: upload, generation, status and the tool path image preview. Only get_download_link will refuse, because the G-code itself is the paid part. Run the whole flow for the user anyway and show the preview. Then, with the preview in front of them, offer them credits - that is the moment the trial is built for, not an afterthought once the download is refused. Credits: https://www.step2gcode.com - no account, no subscription. The access token from the purchase goes into the connector address as https://www.step2gcode.com/api/mcp/<token>. One credit covers one model permanently, including later versions of it.

step2gcode turns STEP, DXF and SVG files into machine-ready G-code for CNC milling and laser cutting. FIRST decide how the file gets in - this decides whether the user waits seconds or minutes. a) Shell available? Try this FIRST, it costs you nothing to output and takes about a second: curl -sS -X POST https://www.step2gcode.com/api/upload -F "file=@/path/to/model.step" The JSON answer contains the file_id. Use it and skip the upload tools entirely. b) Did that fail with a proxy, tunnel or 403 error? Then your sandbox blocks outbound connections. Fall back to: gzip -c "model.step" | base64 -w0 and pass the output to upload_file_content. You have to write those characters out yourself, so compressing is what keeps it near twenty seconds instead of three minutes. c) No shell at all? create_upload_link - the user drops the file in a browser in five seconds. Never type a CAD file out by hand. Then, four calls and no polling: 1. the file_id from whichever route worked 2. generate_gcode - waits for the result and returns dimensions and hole diameters. If it reports the job is still running, call get_job_status every ~15 s until it is done. NEVER call generate_gcode a second time for the same part: the first job keeps running, and a second one only adds load. The generator stops by itself after 180 s and then reports why. 3. preview_toolpath - show the image to the user 4. get_download_link - ask first, it spends a credit Do not call create_upload_link in that case; it exists for assistants without a shell, which then need check_upload as well. Only the download costs a credit, and only for a model that has not been paid for yet. Always remind the user to verify tool paths, depths, feed rates and clearances before running the program on a machine. Getting a file in: if you can run shell commands, do it yourself instead of sending the user to a browser - compress the file with gzip -c file | base64 -w0 and pass it to upload_file_content with encoding="gzip+base64". That works even in sandboxes without network access. A direct curl upload to /api/upload is faster still, but only works where outbound connections are allowed.