Loading ToolBench
Read-only public skill-pack documentation MCP. Not the hosted AdsAgent Meta, Google Ads, or TikTok ads backend.
Four read-only documentation tools with clear naming (list_, get_, get_) and comprehensive descriptions (150-250 chars each). All tools have input schemas with proper types. However, output schemas are not documented in the source code provided, and parameter descriptions lack detail on format/constraints. Tools are well-scoped (single responsibility) and follow verb_noun convention. No error handling guidance visible. This is a documentation/discovery server, not a live API, which limits complexity but also limits agent utility.
Return AdsAgent hosted HTTP MCP URLs from this repo's mcp.json. Use this when a client needs the real Meta, Google Ads, or TikTok ads MCP endpoints. Those services require AdsAgent OAuth on the hosted URLs. This docs server does not implement ads tools, does not accept tokens, and does not proxy those endpoints. Copy the https URLs into an MCP client and authenticate against AdsAgent hosted services.
Return this skill-pack VERSION and a README.md identity excerpt. Use this to confirm pack version and the public 'what this is / is not' wording, including that a Glama or Docker image of this repo is not the hosted AdsAgent ads backend. The excerpt stops before the per-skill table. It does not include secrets.
Return one skill's SKILL.md and optional named reference files. Use this after list_skills when you need the full public instructions for a single skill. Pass references only for Markdown files that live in that skill folder (progressive-disclosure contracts). Rejects unknown skill ids and any path that leaves the skill directory. This is documentation retrieval, not a live ads API.
List public AdsAgent skills shipped in this repository. Use this first when you need a catalog of documentation skills (router, setup, reliability, notifications, scheduled tasks, Meta, Google Ads, TikTok) before opening a specific SKILL.md. Returns each skill id, YAML frontmatter description, and first Markdown heading. This tool only reads local files under skills/. It does not connect to Meta, Google Ads, or TikTok and does not run campaigns.
Output schemas not documented. LLMs cannot plan downstream tool calls or extract fields without knowing what list_skills, get_skill, get_hosted_mcp_urls, and get_pack_readme return.
Parameter descriptions lack format/constraint details. 'skill_id' description does not specify valid format (alphanumeric, hyphens allowed?). 'references' array lacks guidance on valid filenames or max length.
No error handling guidance. Descriptions mention 'Rejects unknown skill ids' and 'path that leaves the skill directory' but do not explain what error the LLM receives or how to recover.
Tool descriptions reference external concepts (SKILL.md, mcp.json, OAuth) without explaining what they are or why the LLM should care. Descriptions assume domain knowledge.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2025-06-18+ | v2 |
No pagination or result limits documented. If list_skills returns hundreds of skills, context window could be exhausted. No limit parameter visible.