Loading ToolBench
Official Model Context Protocol server for the Swfte AI platform — agents, chatflows, workflows, RAG, voice, marketplace. Build, validate, and deploy complete workflows, agents, chatflows, widgets, applications and MCP servers from Claude Code, Claude Desktop, Cursor, or any MCP client.
Strong naming conventions (verb_noun pattern throughout) and comprehensive parameter schemas with type definitions. All 24 tools have descriptions and input schemas. However, descriptions are often terse (many under 100 chars) and lack actionable guidance for LLM selection. Output schemas are not documented. Error handling guidance is minimal. Tool composition is sound but some tools could be consolidated (e.g., multiple status/preview/files tools for the same resource).
Abort an AGENT-mode build for an AppWizard session.
Delete a saved hosted app artifact. Stop/destroy its deployed capacity separately.
Get a short-lived download URL for a saved hosted app artifact.
Fork an existing hosted app artifact into the authenticated workspace.
Read a saved hosted app artifact.
List saved hosted app artifacts for the authenticated workspace.
Package and save generated artifacts for an AppWizard session.
Output schemas not documented. Tools return data but LLMs cannot predict field structure, forcing exploratory calls and wasting tokens.
Descriptions are terse (many 40-80 chars) and lack actionable guidance. LLMs cannot distinguish when to call swfte_app_wizard_status vs swfte_app_wizard_plan vs swfte_app_wizard_preview, all read session state but return different aspects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Create a hosted AppWizard build session (billable container capacity). Distinct from swfte_build kind application, which creates a blueprint. Poll status; retain sessionId for cleanup. Optional designContext injects referenced cases and explicit product/workflow/agentic guidance.
Deploy a completed hosted app artifact. Provisions billable capacity; inspect deployment status/logs afterwards.
Destroy a hosted app deployment and release its resources.
Read hosted app deployment status without assuming readiness.
Read hosted app deployment logs.
Redeploy a hosted app. Provisions billable capacity.
Stop hosted app runtime capacity. Cleanup remains available when deployment is disabled.
List hosted app deployments for the authenticated workspace.
Destroy an AppWizard session and its build droplets. Hosted deployments have a separate destroy tool.
Read a generated source file using a relative path.
Read AppWizard session files. Status is returned verbatim; a preview URL alone does not prove build completion.
Read AppWizard session plan. Status is returned verbatim; a preview URL alone does not prove build completion.
Read AppWizard session preview. Status is returned verbatim; a preview URL alone does not prove build completion.
Send a follow-up prompt to an existing HUMAN-mode AppWizard session. Do not repeat on timeout; inspect status.
Restore an AppWizard session from its saved S3 snapshot; overwrites workspace files.
Save a recoverable workspace snapshot for an AppWizard session.
Read AppWizard session status. Status is returned verbatim; a preview URL alone does not prove build completion.
No error recovery guidance. Descriptions state what tools do but not what to do if they fail. E.g., swfte_app_wizard_create says 'Poll status' but does not explain timeout behavior or retry strategy.
Destructive operations (destroy, delete) lack confirmation or dry-run support. Agents can irreversibly delete artifacts or deployments without explicit user consent.
Tool composition: swfte_app_wizard_status, swfte_app_wizard_plan, swfte_app_wizard_preview, and swfte_app_wizard_files all read the same session but return different aspects. Consider consolidating into a single swfte_app_wizard_get_session with optional fields parameter.