Loading ToolBench
MCP server for BlogBurst — your Virtual CMO. Autonomous AI marketing agent: auto-post to Twitter/X, Bluesky, Telegram, Discord, auto-engage (replies, likes, follows), SEO/GEO audits, competitor intelligence, community scanner, growth diagnostic, task management. Free tier available.
BlogBurst provides 12 tools with mostly complete schemas and descriptions. However, there are significant gaps in output schema documentation, missing parameter descriptions for optional fields, and weak error handling guidance. The tools follow verb_noun naming conventions well, but descriptions could be more detailed about when to use each tool versus alternatives. Tool composition is reasonable but lacks dependency documentation (e.g., which tools must be called before others). No tool annotations (readOnlyHint/destructiveHint) are present despite clear read vs. write distinctions. Descriptions range from 90 - 280 chars, mostly within acceptable bounds. Schemas are present for all tools with proper enums and type definitions, but output schemas are completely undocumented, LLMs cannot predict what fields to expect in responses.
Chat with your AI marketing agent. It can generate content, check analytics, manage auto-pilot, view trending topics, and more — all through natural conversation. This is the most powerful tool — use it for any complex or multi-step marketing task.
AI-powered title brainstorming via multi-turn conversation. Send messages to iteratively refine blog post titles.
Configure the autonomous posting agent. Enable/disable auto-pilot, set posting frequency, choose platforms, and set timezone.
Generate a complete AI-written blog post from a topic. Returns title, content (markdown), meta description, and tags.
Generate social media posts directly from a topic (no blog needed). Creates platform-optimized content with correct character limits.
NO OUTPUT SCHEMAS DOCUMENTED: All 12 tools lack return type documentation. LLMs cannot predict response structure, fields, or data types. This forces agents to discover via trial-and-error and wastes context windows parsing unexpected results.
MISSING TOOL ANNOTATIONS: No destructiveHint or readOnlyHint attributes on tools despite clear semantic distinctions. publish_content, configure_auto_pilot, and run_auto_pilot_now are write/destructive but not marked. LLMs cannot distinguish safe vs. irreversible operations without explicit hints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 59 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Check the status of your autonomous posting agent (auto-pilot). Shows if enabled, which platforms, posts per day, timezone, and last run time.
Check which social media platforms are connected to your BlogBurst account and ready for auto-publishing.
Discover trending topics from HackerNews, Reddit, Google Trends, and Product Hunt. Great for finding content inspiration.
Check your current API usage, remaining generations, and plan limits.
Publish content directly to your connected social media platforms (Twitter/X, Bluesky, Telegram, Discord). Requires platforms to be connected first at https://blogburst.ai/dashboard/connections — Twitter/X is one-click OAuth (5 seconds).
Turn a blog URL or raw text into ready-to-post social media content for multiple platforms. Supports Twitter, LinkedIn, Reddit, Bluesky, Threads, Telegram, Discord, TikTok, YouTube.
Trigger the auto-pilot to generate and publish content immediately, without waiting for the next scheduled run.
WEAK DESCRIPTIONS FOR UTILITY TOOLS: get_connected_platforms (50 chars), get_usage (50 chars), get_auto_pilot_status (65 chars), and run_auto_pilot_now (60 chars) have truncated descriptions that provide minimal context. LLMs cannot determine WHEN to call these vs. alternatives. Descriptions should be 100 - 200 chars explaining purpose and use case.
MISSING PARAMETER DESCRIPTIONS FOR OPTIONAL FIELDS: In get_trending_topics, parameters niche, source, and limit lack 'required' declarations. LLMs cannot infer whether these are mandatory or optional. All parameters must explicitly state required vs. optional in both schema and description.
NO ERROR HANDLING OR RECOVERY GUIDANCE: None of the 12 tools provide error messages, recovery hints, or actionable next steps. If publish_content fails (e.g., platform not connected), the agent has no guidance on how to proceed. Tools should document what can fail and suggest corrective actions (e.g., 'Call get_connected_platforms() first').
NO PAGINATION OR RESULT LIMITS DECLARED: generate_platform_content, repurpose_content, and get_trending_topics could return large result sets (multiple platform variants, multiple topics). Descriptions do not mention pagination, limit defaults, or maximum result counts. LLMs risk context window exhaustion if a tool returns 100+ items.
MISSING DEPENDENCY DOCUMENTATION: brainstorm_titles and generate_blog could benefit from multi-step workflows, but there are no hints about prerequisite calls. For example, agent_chat is described as 'the most powerful tool', this is vague guidance that forces LLMs to reason about when to use it vs. single-purpose tools. Tool descriptions should state dependency chains (e.g., 'Call get_trending_topics() first to discover ideas, then pass topics to generate_blog()').
VAGUE TOOL SEMANTICS: agent_chat is described as capable of 'generate content, check analytics, manage auto-pilot, view trending topics, and more, all through natural conversation.' This is a catch-all tool that overlaps with dedicated tools. LLMs will use agent_chat indiscriminately, bypassing structured tools. Either split agent_chat into focused tools or document exactly when to use it vs. specific tools like generate_blog or configure_auto_pilot.
NO CONFIRMATION PATTERN FOR DESTRUCTIVE OPS: publish_content, configure_auto_pilot, and run_auto_pilot_now lack confirmation/dry-run steps. An LLM could accidentally publish spam to all connected platforms or enable auto-pilot with wrong settings. Irreversible operations should support a dry-run or explicit confirmation pattern.
IMAGE_URLS PARAMETER LACKS VALIDATION RULES: publish_content accepts image_urls as an array, but there is no description of valid formats (https only? max count? max file size?). This invites invalid or malicious inputs.