Loading ToolBench
MCP server that integrates with the Graphlit Platform for LLM-enabled knowledge API operations, supporting content ingestion, retrieval, and management from multiple sources including Slack, Discord, Notion, Google Drive, email, Jira, Linear, and GitHub.
Three tools with mixed quality. Tool names lack action verbs (configureProject, askGraphlit, retrieveSources are weak). Descriptions are present but verbose (200+ chars) and lack clear WHEN/WHY guidance. Schemas use Zod with types and descriptions, but parameter descriptions are generic. retrieveSources has 6 params with optional filters, good for flexibility but risks LLM confusion on when to use vs queryContents. No output schema documented. Error handling minimal (try/catch returns JSON string, no recovery guidance). configureProject modifies state but description doesn't explicitly warn of side effects.
Ask questions about the Graphlit API or SDKs. Can create code samples for any API call. Accepts an LLM user prompt for code generation. Returns the LLM prompt completion in Markdown format.
Configures the default content workflow for the Graphlit project. Only needed if user asks to configure the default workflow. Optionally accepts whether to enable high-quality document and web page preparation using a vision LLM. Defaults to using Azure AI Document Intelligence for document preparation, if not assigned. Optionally accepts whether to enable entity extraction using LLM into the knowledge graph. Defaults to no entity extraction, if not assigned. Optionally accepts the preferred model provider service type, i.e. Anthropic, OpenAI, Google. Defaults to Anthropic if not provided. Returns the project identifier.
Retrieve relevant content sources from Graphlit knowledge base. Do *not* use for retrieving content by content identifier - retrieve content resource instead, with URI 'contents://{id}'. Accepts an LLM user prompt for content retrieval. For best retrieval quality, provide only key words or phrases from the user prompt, which will be used to create text embeddings for a vector search query. Only use when there is a valid LLM user prompt for content retrieval, otherwise use queryContents. For example 'recent content' is not a useful user prompt, since it doesn't reference the text in the content. Accepts an optional ingestion recency filter (defaults to null, meaning all time), and optional content type and file type filters. Also accepts optional feed and collection identifiers to filter content by. Returns the ranked content sources, including their content resource URI to retrieve the complete Markdown text.
Tool names lack action verbs. 'configureProject' should be 'configure_project' or 'update_project_workflow'. 'askGraphlit' is vague, 'query_api_docs' or 'generate_code_sample' is clearer. 'retrieveSources' should be 'search_sources' or 'query_knowledge_base'. LLMs infer intent from names before reading descriptions.
No output schemas documented. LLMs cannot plan downstream calls or extract fields. configureProject returns JSON string of {id}, askGraphlit returns Markdown, retrieveSources returns ranked sources, all undocumented. Agents must guess structure.
configureProject description does not warn that it modifies project state (creates specifications, workflows, updates project). Agents need to know this is irreversible. Missing 'This tool modifies your project configuration' warning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
retrieveSources has 6 optional parameters (inLast, contentType, fileType, feeds, collections) but description does not explain when to use this tool vs queryContents. Guidance missing: 'Use when you have a text query; use queryContents for ID-based retrieval.' LLM may pick wrong tool.
Error handling returns raw error messages as JSON strings with isError flag. No recovery guidance. 'Error: Invalid model service type' tells agent nothing, should be 'Invalid model service type. Supported: Anthropic, OpenAI, Google. Please retry with a valid type.'