twocents MCP - Human Feedback Loop for AI Agents¶

twocents gives AI agents a direct channel to request human feedback. Agents can share generated content, designs, or proposals via a simple web page, and humans can leave structured feedback that flows back into the agent's context.

What It Does¶

Share pages - Agent creates a simple, shareable web page with content for review
Collect feedback - Humans leave comments, approvals, or revision requests
Structured feedback - Feedback can be structured (approve/reject/revise) or freeform
Remote MCP - Works as a remote endpoint, no local install needed

Quick Start¶

# Remote endpoint - no local install
hermes mcp add twocents --url https://twocents.page/mcp

# Or Claude Code
claude mcp add twocents --url https://twocents.page/mcp

Prerequisites: None. The endpoint is open and requires no authentication.

Key Tools¶

Tool	Description
`share_page`	Create a shareable page with content for human review
`get_feedback`	Retrieve all feedback left on a shared page
`get_page_status`	Check if a page has been reviewed
`list_pages`	List all pages shared by this agent

Use Cases¶

Content review - Agent drafts a blog post, shares it via twocents, PM reviews and approves
Design feedback - Agent generates UI mockups → shares → designer leaves pixel-level feedback
Decision gates - Agent proposes architecture decision → shares for team vote → proceeds on approval
Client review - Agent drafts deliverables → shares with client → incorporates feedback in next iteration

Hermes Agent Integration¶

# Agent creates a review page
page = mcp_twocents_share_page(
    title="Q3 Content Calendar Draft",
    content=calendar_markdown,
    feedback_type="approve_or_revise"
)
# Later: check for feedback
feedback = mcp_twocents_get_feedback(page_id=page.id)
if feedback.status == "approved":
    print("Calendar approved - proceeding with scheduling")

Discovered via mcp.so sweep. Listed in the CorpusIQ MCP Catalog.