Loading ToolBench
DEPRECATED — the official GEN MCP is the hosted server at https://mcp.gen.pro. MCP tools for the GEN autonomous social media agent platform, providing programmatic access to agent setup, content idea generation, vidsheet management, and video production.
Server has 13 tools with consistent naming (verb_noun pattern: gen_get_*, gen_list_*, gen_create_*, gen_update_*, gen_delete_*). All tools have descriptions (10 - 200 chars, within baseline 34 - 392 range). Input schemas are visible and properly typed with Annotated/Field patterns. However, descriptions are minimal (often under 100 chars) and lack actionable context for LLM selection. Parameter descriptions are present but generic. Output schemas are not documented, responses are JSON strings with no declared structure. Error handling is absent from visible code. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY, WRITE, DESTRUCTIVE). Missing guidance on when to call each tool vs. similar ones (e.g., gen_list_organizations vs. gen_list_workspaces). Composition is sound, tools chain via IDs, but descriptions don't guide the agent through the setup workflow.
Step 1 (Agent Setup): Create a new agent inside a workspace. After creation, use gen_update_agent_core to fill in identity, overview, personality, voice, and inspiration sources.
Step 1 (Agent Setup): Create a new organization/workspace. You become owner automatically. Only needed if the user doesn't already have a workspace.
Step 1 (Agent Setup): Soft-delete an agent (requires owner/manager role or being the creator).
Step 1 (Agent Setup): Permanently delete an organization and all associated data (requires owner role, irreversible).
Step 1 (Agent Setup): Get full details of a specific agent by ID. For reading full setup state (identity + overview + personality + voice + inspiration + accounts), prefer gen_get_agent_core.
Output schemas not documented. Tools return json_result(data) with no declared structure. LLMs cannot plan downstream calls or extract required fields without trial-and-error.
Tool annotations missing. Risk levels (READ_ONLY, WRITE, DESTRUCTIVE) are documented in metadata but not exposed via MCP readOnlyHint/destructiveHint/idempotentHint. Agents cannot infer safety/retry semantics.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Step 1 (Agent Setup): Get the authenticated user's profile and workspace memberships. Always call first to verify your PAT and discover which workspaces you have access to.
Step 1 (Agent Setup): Get details of a specific organization by ID.
Step 1 (Agent Setup): List avatar images for an agent, with the primary avatar first.
Step 1 (Agent Setup): List agents, optionally filtered by workspace. Use the returned agent_id for ALL downstream content operations (ideas, vidsheets, generations).
Step 1 (Agent Setup): List all organizations/workspaces the authenticated user is a member of, including credits, role, and plan.
Step 1 (Agent Setup): List all workspaces the authenticated user has access to. A workspace is the billing container; every agent lives inside one.
Step 1 (Agent Setup): Update an existing agent's name, description, time zone, or voice keys. For richer setup updates (personality, inspiration, accounts, voice binding), use gen_update_agent_core.
Step 1 (Agent Setup): Update an organization's name (requires owner or manager role).
Descriptions lack actionable context. 'List all organizations/workspaces...' does not explain WHEN to call this vs. gen_list_workspaces, or what the response structure is. Baseline for A+ tools is 50 - 200 chars with clear selection criteria.
No error handling guidance. Code does not show try/catch, validation, or recovery hints. If an API call fails, the agent receives a raw error with no actionable next step.
Pagination not visible for list tools. gen_list_agents, gen_list_organizations, gen_list_workspaces, gen_list_agent_avatars accept no limit/offset/cursor params (except gen_list_agent_avatars has cursor). Large result sets will blow context window.