Loading ToolBench
Official Riven MCP server — exposes Riven chat, research, council, and usage tools over stdio. Provides access to 75+ AI models with transparent per-token pricing, intelligent routing, usage tracking, and on-prem model health monitoring.
Mixed implementation across TypeScript and Python components. TypeScript tools (riven_chat, riven_research, riven_council, riven_usage) have good descriptions but lack visible input schemas in the provided code. Python tools (list_models through get_model_pricing) are referenced but source code not provided, making schema verification impossible. Tool naming follows verb_noun convention well. Critical gap: cannot verify input schemas for any tool from provided source. Descriptions range 50-150 chars, meeting baseline. No evidence of output schema documentation, error recovery guidance, or parameter validation rules.
Send a chat completion request to a Riven model with support for system prompts, temperature, max_tokens, and other OpenAI-compatible parameters.
Check the health and availability status of on-prem models (e.g. GLM-5.2 on A100 GPU) and cloud model providers.
Compare multiple models side-by-side on pricing, latency, context length, and capabilities to help select the best model for a task.
Get detailed pricing information for a specific model, including per-token input/output costs, cached input rates, and volume discounts.
Retrieve token usage and cost breakdowns for the authenticated account, with optional filtering by date range and grouping by model or day.
List all available models on the Riven platform with pricing, capabilities, and provider information.
Input schemas not visible in provided source code for 10 of 12 tools. Cannot verify parameter types, descriptions, constraints, or defaults.
No output schema documentation visible. Tools return data but LLMs cannot plan downstream calls without knowing response structure (fields, types, pagination). Violates pattern:response-shaper.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
Send a single prompt to a Riven chat model and return the completion text. Use for quick, single-turn questions that don't need agentic research.
Create a Riven Computer council task, which asks multiple models to answer the same prompt and produces a synthesized verdict plus each model's individual answer. Use when you want a cross-checked or debated answer rather than a single model's view.
Create a Riven Computer research task for the given prompt, run it, poll until it completes (or times out), and return a summary of the resulting thread. Use for open-ended research questions that benefit from multi-step agentic work.
Return current usage statistics for the authenticated Riven account.
Use Riven's intelligent model routing to auto-select the best model based on cost, latency, and capability constraints, with automatic fallback if the primary model fails.
Submit URLs to Bing IndexNow for immediate indexing. IndexNow is a protocol that notifies search engines of content changes, enabling faster crawling and indexing.
No error recovery guidance visible. Tools lack descriptions of failure modes, retryability, or actionable error messages. Agents cannot self-correct on failures.
route_request and submit_indexnow have partial schemas visible but lack descriptions for several parameters (e.g., 'strategy' enum values not documented, 'host' override behavior unclear). Parameter descriptions must explain constraints and valid values.
Python tool implementations referenced but source code not provided (python/src/riven_mcp/tools/*.py files). Cannot verify schema completeness, parameter validation, or error handling for 8 tools.