Loading ToolBench
Agent-agnostic MCP server exposing reusable database, PDF, and workflow skills as Tools and Resources.
Server has 32 tools with consistent naming (verb_noun pattern) and descriptions present for all. However, many descriptions are generic or lack actionable context. Input schemas are visible and typed, but parameter descriptions are often minimal. Output schemas are not documented. Error handling is basic. Tool composition is reasonable but some tools lack clear prerequisites or dependencies. Security considerations (db_write confirmation, deploy preflight) are present but inconsistently applied.
Bump the version number in package.json and create a git tag
Check the status of the continuous integration/deployment pipeline
Check for available updates to the mcp-skills server
Execute a read-only SQL query against a configured database connection
Execute a write SQL query (INSERT, UPDATE, DELETE) against a configured database connection with confirmation gate for production
Deploy the application to production with preflight checks
Get boot diagnostics and health information for the server
Get the canonical triage workflow for error patterns and health diagnostics
Output schemas not documented. Tools return results but LLMs cannot predict field structure for downstream chaining. E.g., db_read, list_db_connections, summarize_mcp_errors lack documented return types.
Parameter descriptions are minimal or generic. E.g., 'page_ranges' in pdf_split is described as 'Comma-separated page ranges (e.g., 1-3,5,7-9)' but lacks validation rules, min/max, or error guidance. LLMs cannot infer constraints.
Error handling guidance missing. Tools like db_write, deploy, and playwright_execute are high-risk but lack recovery hints. E.g., db_write says 'confirmation gate for production' but does not explain what happens if confirmation fails or how to retry.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | <=2025-11-25 | v2 |
Get the canonical Playwright browser automation workflow for authenticated multi-step verification
Get the version of the mcp-skills server
Get the canonical worktree-proposal workflow for context switches and source changes
List available database connections
Introspection for the MCP_SKILLS_SELECT tool-slicing system. Reports the available groups and their tools, the always-on tools, the active selector, any unrecognized selector tokens, and the tool names that actually registered in this process. Available in every slice.
Mark an error pattern as resolved after a fix is deployed
Update the status of a pattern note (active, superseded, wrong, stale)
Decrypt a password-protected PDF file
Encrypt a PDF file with a password
Extract text content from a PDF file
Merge multiple PDF files into a single document
Rotate pages in a PDF file
Split a PDF file into separate pages or page ranges
Add a watermark to PDF pages
Close a Playwright browser session
Execute Playwright commands in an active browser session
Prepare a Playwright browser session for a named target with authentication
List active Playwright browser sessions
Start recording audio from the system microphone
Record a triage note for an error pattern to guide future agents
Save a journal entry to a durable notes store
Save a session note to a durable notes store
Stop the current audio recording
Summarize and analyze MCP tool error patterns from the audit log
Tool composition gaps: playwright_prepare requires 'session_id' but does not document how to generate one. playwright_execute depends on playwright_prepare but the dependency is not stated in descriptions. Agents may call execute before prepare.
Irreversible operations lack dry-run or confirmation patterns. deploy and db_write have partial guards (preflight, confirm token) but no dry-run mode. Agents cannot preview consequences before committing.