Start a Ralph-On-Shelf autonomous loop in an E2B sandbox with MCP tools
Launches an autonomous Claude agent in an E2B sandbox with optional MCP server integration for web search and AWS knowledge.
/plugin marketplace add HarleyCoops/RalphOnAShelf/plugin install ralph-on-shelf@ralph-on-shelf-marketplaceLaunch an autonomous Claude agent inside an E2B cloud sandbox with MCP connector support.
When mcp=auto (default), Ralph analyzes your prompt and automatically connects to relevant MCP servers:
Use mcp=none to disable MCP and use the classic SDK mode.
Use mcp=exa-search,aws-knowledge to explicitly specify servers.
Run the following Python code to start the Ralph loop:
cd {{cwd}} && python -c "
from main import launch_ralph
result = launch_ralph(
prompt='''{{prompt}}''',
max_iterations={{max-iterations | default: 10}},
completion_promise='{{completion-promise | default: COMPLETE}}',
output_dir='{{output-dir | default: ralph_output}}',
mcp='{{mcp | default: auto}}',
timeout=600
)
print(f'[ralph] final status: {result[\"status\"]}')
print(f'[ralph] total iterations: {result[\"iterations\"]}/{result[\"max_iterations\"]}')
print(f'[ralph] MCP servers: {result.get(\"mcp_servers\", [])}')
if result.get('downloaded_files'):
print(f'[ralph] files saved to: {result[\"output_dir\"]}')
for f in result['downloaded_files']:
print(f' - {f}')
"
The agent will run autonomously in the E2B sandbox until: