Skip to content

Proposal: Stream Tool Arguments #82

Open
Open
Proposal: Stream Tool Arguments#82
Labels
enhancementNew feature or request

Description

@MiguelsPizza
MiguelsPizza
opened on Feb 13, 2026

Currently, execute fires after the full args JSON is parsed. For tools with large text args (email bodies, code), it'd be nice to stream chunks into page UI as the model generates them. example from userland

navigator.modelContext.registerTool({
  name: "compose_email",
  stream: true,
  inputSchema: { /* to, subject, body */ },
  execute: async (call) => {
    composer.open();
    for await (const { key, chunk } of call) {
      if (key === "to")   composer.appendTo(chunk);
      if (key === "body") composer.appendBody(chunk);
    }
    const { to, subject, body } = await call.args;
    return { content: [{ type: "text", text: `Draft created` }] };
  }
});

For declarative tools this could be default behavior.

Activity

  1. Imageanssiko
    added
    Agenda+
    on Feb 13, 2026
  2. Imageanssiko
    removed
    Agenda+
    on Feb 27, 2026

  3. MiguelsPizza commented on Mar 9, 2026

    @MiguelsPizza
    MiguelsPizza
    on Mar 9, 2026
    ContributorAuthor

    Here's a streaming demo running in a local Chromium fork with the proposed API:

    WebMCP.stream.demo.mp4 
    navigator.modelContext.registerTool({
    name: 'compose_site',                                                                                                                
    description: 'Stream HTML progressively into a live preview.',                                                                       
    stream: true,                                                                                                                        
    inputSchema: {                                                                                                                       
      type: 'object',                                                                                                                    
      properties: { html: { type: 'string' } },                   
      required: ['html'],
    },
    async execute(call) {
      for await (const { key, chunk } of call) {                                                                                         
        if (key === 'html') {
          htmlBuffer += chunk;                                                                                                           
          preview.innerHTML = htmlBuffer;                         
        }
      }                                                                                                                                  
   
      const { html } = await call.args;                                                                                                  
      preview.innerHTML = html;                                   
      return { content: [{ type: 'text', text: 'Site composed.' }] };                                                                    
    },                                                                                                                                   
  });

    I think this API would be a great way to show off what you can make with WebMCP and how it is different from MCP.

  4. Imagemarkafoltz
    added
    enhancementNew feature or request
    on Sep 24, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions