Loading ToolBench
Create custom MCP servers with AI assistance
This server exhibits significant definition quality gaps across its 38 tools. While naming conventions are generally acceptable (verb-first patterns like 'get', 'create', 'list', 'update'), descriptions are inconsistent and often too brief (many under 50 chars). Input schemas are present but sparse, most parameters lack proper type definitions or detailed constraints. Output schemas are undocumented. The server mixes infrastructure tools (listServers, createMcpServer) with domain-specific tools (BigQuery, Jira, PostgreSQL, weather), which is architecturally questionable. Error handling and recovery guidance are absent from all visible tool definitions. Security concerns exist: tool descriptions do not address authentication or credential handling (especially for BigQuery, Jira, PostgreSQL). No evidence of per-tool permission scopes or audit logging.
Analyze a server code to identify required npm dependencies
Click an element on the page using a CSS selector
Close the browser instance
Create a new dataset in BigQuery
Create a new issue in Jira
Create a new MCP server, save it to disk, and register it with Claude Desktop
Create a subtask for a parent issue in Jira
Critical: No output schemas documented for ANY tool. LLMs cannot plan downstream calls or extract necessary fields without knowing response structure. This violates pattern:tool and mxe:include-chaining-ids.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 43 | - | v1 |
Delete a table from BigQuery
Get schema information for a PostgreSQL table
Execute a SQL query against a PostgreSQL database
Execute arbitrary JavaScript code in the browser context
Execute a SQL query against BigQuery
Execute a query and export results to a BigQuery table
Extract elements from the page using a CSS selector
Get the current Claude Desktop configuration
Fetch current weather data for a location
Get detailed information about a Jira issue
Get the visible text content of the current page
Retrieve the content of an existing server file with line numbers
Get the schema of a BigQuery table
Get an example MCP server template structure
Fetch weather forecast for a location
Initialize a Puppeteer browser instance
Install npm dependencies for an MCP server
Create a link between two Jira issues
List all datasets in a Google Cloud project
List all Jira projects
List available MCP servers
List all tables in a BigQuery dataset
Navigate to a URL in the browser
Test run an MCP server directly to debug issues
Search for a city to get the correct name format for weather queries
Search for Jira issues using JQL
Take a screenshot of the current page
Test the PostgreSQL database connection
Type text into an input element
Update the Claude Desktop configuration to register an MCP server
Update an existing MCP server with new code
High: Destructive/state-modifying tools (createMcpServer, updateMcpServer, deleteTable, createIssue, updateClaudeConfig) lack confirmation or dry-run capability. Agents may accidentally delete data without recourse. Violates pattern:confirmation-request.
High: Security concern, authentication and credential handling not addressed in any BigQuery, Jira, or PostgreSQL tool descriptions. No mention of required API keys, environment variables, or permissions. Violates pattern:secret-injection and pattern:scope-declaration.
High: No error guidance in tool descriptions. Errors return no actionable recovery steps (e.g., 'Query failed' with no hint about syntax, quota, or permissions). Violates pattern:recovery-guide.
Medium: Many tools lack description detail on constraints, formats, and valid values. E.g., 'updateType' enum is documented in schema but description is bare. 'location' for getCurrentWeather lacks format spec ('London,UK or New York,US' is in description but not as a format constraint). Violates pattern:constrained-input.
Medium: Pagination support missing or undocumented. Tools like 'listDatasets', 'listTables', 'listProjects', 'searchIssues' do not document limit/offset or whether results are capped. LLMs may expect unbounded lists and hit context window limits. Violates pattern:paginated-result.
Medium: Server architecture is questionable, mixing infrastructure/meta tools (createMcpServer, updateMcpServer, getClaudeConfig) with domain-specific API wrappers (BigQuery, Jira, PostgreSQL, weather). This violates pattern:tool composition; a server should have a clear, singular responsibility. Recommend splitting into separate, focused servers.
Low: Some tool names could be more specific. 'getServerContent' is vague, does it return code, metadata, or both? 'runServerDirectly' is unclear about output (logs, errors, success/failure). Violates pattern:tool naming clarity.