Understand this story
OpenRouter announced that its MCP has received a major upgrade: users can now directly call 600+ models, the Decisions API, and image, speech, and video generation from within agents without writing code first. The context injected by model queries has been reduced from about 1 MB (roughly 250k tokens) to 38 KB with pagination, a 96% reduction; video generation has been changed to asynchronous tasks, with results returned as links. OpenRouter says that in its tests Claude Code improved from 5/8 tasks to 8/8, @manufact gave it a score of 100/100, and plugin stores for OpenAI, Anthropic, and others will be added later.
AI-generated from public reports
Developments
1 confirmed developments
- Source:OpenRouter
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 10
OpenRouterFirst-party
OpenRouter MCP Upgrade: Model Query Context Reduced by 96%, Installation Can Be Completed by Agents Reading It Themselves
OpenRouter announced an MCP upgrade, allowing users to directly call 600+ models, the Decisions API, and image, speech, and video generation within agent conversations. Model query responses have been reduced from about 1 MB (about 250k tokens) to 38 KB with pagination, a 96% reduction; video tasks are initiated by the agent and then retrieve a link, no longer stuffing raw bytes into the context. The installation method has changed to having the agent read the installation page and complete it on its own; the official statement says that in its tests, Claude Code improved from 5/8 tasks to 8/8.
OpenRouterFirst-party
OpenRouter MCP Upgrade and Coming Soon to Plugin Stores
OpenRouter announced that MCP has completed a major upgrade and will soon be available in plugin stores such as OpenAI and Anthropic, while also supporting long-term persistent login. After the upgrade, the volume of model query results injected into the agent context dropped from about 1 MB (about 250k tokens) to 38 KB with pagination, a 96% reduction; video generation now returns links in a task-based manner instead of stuffing raw bytes into the context. The company says that in its own tests, Claude Code's task completion count rose from 5/8 to 8/8, @manufact gave it a score of 100/100, and users can also have the agent read the installation page and complete configuration on its own.
OpenRouterFirst-party
OpenRouter MCP now supports viewing usage and spending within conversations
OpenRouter announced that after its MCP upgrade, agents can directly display the current key, quota limits, and the costs incurred in this session within the conversation, without leaving the conversation.
OpenRouterFirst-party
OpenRouter MCP Upgrade: Claude Code Task Completion Rate Rises from 5/8 to 8/8
OpenRouter upgraded MCP to improve the agent tool-calling experience (AgEx), making tools easier for agents to understand and select, providing structured output when needed, and having the server tell Claude Code and Codex when to call. The company says that in its own tests, Claude Code improved from 5/8 tasks to 8/8, and @manufact gave it a score of 100/100. The volume of context injected by model queries dropped from about 1 MB (about 250k tokens) to 38 KB with pagination, a 96% reduction; video generation was changed to an asynchronous task, with results returned as links instead of stuffing raw bytes into the context.
OpenRouterFirst-party
OpenRouter MCP video generation now returns links asynchronously
OpenRouter says its video generation capability on MCP has been enhanced: after an agent initiates a task, it can come back to retrieve the result when the task is complete, and the result is returned as a link rather than stuffing raw bytes into the context.
OpenRouterFirst-party
OpenRouter MCP Upgrade: Model Query Context Reduced by 96%
OpenRouter MCP has received a major upgrade. Model queries previously wrote about 1 MB (roughly 250k tokens) into the agent context; now they are returned in 38 KB paginated form, a 96% reduction in size. Transcript text carries timestamps and speaker information in plain text form, making it convenient for various clients to use.
OpenRouterFirst-party
OpenRouter MCP Upgrade: Model Query Context Reduced by 96%, Claude…
OpenRouter announced that its MCP has undergone a major upgrade, allowing users to have agents directly call 600+ models, the Decisions API, and image, speech, and video generation in chat without writing code first. The context used for model queries has been reduced from about 1 MB (about 250k tokens) to 38 KB and returned in pages, a 96% reduction; video generation has been changed to asynchronous tasks, with results returned as links. OpenRouter says Claude Code in its tests improved from 5/8 tasks to 8/8, @manufact gave it a 100/100 score, and it will later be listed on plugin stores such as OpenAI and Anthropic.
OpenRouterFirst-party
OpenRouter MCP Upgrade: Model Query Context Reduced by 96%, Claude Code Task Completion Rate Rises to 8/8
OpenRouter announced a major upgrade to MCP, allowing users to directly call 600+ models, the Decisions API, and image, speech, and video generation within agents without writing code first. The context injected by model queries has been reduced from about 1 MB (roughly 250k tokens) to 38 KB with pagination, a 96% reduction; video generation has been changed to asynchronous tasks, with results returned as links. The company says Claude Code improved from 5/8 tasks to 8/8 in its tests, @manufact gave it a score of 100/100, and it will subsequently be listed on plugin stores such as OpenAI and Anthropic.