@unihodl/mcp-server
UNIHODL — Agent Handoff SDK & MCP Server
The handoff layer for the agentic web. Capture a human's working session — open tabs, scroll positions, video timestamps, and reasoning thread — and hand it to any AI agent as a signed, scoped, revocable Resume Token. Your agent picks up exactly where the human left off instead of starting cold.
This is the open-source SDK + MCP server + protocol spec. The browser extension, web app, and product live separately at unihodl.app.
# Zero-setup demo — no signup, sandbox key, real API:
npx -y @unihodl/mcp-server
# env: UNIHODL_API_KEY=uh_test_sandbox_demo_key_v0
# then ask your agent: resume ses_8f3aZ91bWhat's here
Package | What it is | Install |
MCP server — gives Claude Desktop, Cursor, Cline, etc. a |
| |
TypeScript client — mint tokens, hydrate sessions |
| |
Python client |
| |
The open Resume Token protocol spec | — |
Related MCP server: Remote Browser
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"unihodl": {
"command": "npx",
"args": ["-y", "@unihodl/mcp-server"],
"env": { "UNIHODL_API_KEY": "uh_test_sandbox_demo_key_v0" }
}
}
}Restart Claude. The resume and list_sessions tools appear. Ask it to
resume ses_8f3aZ91b and it receives the human's conclusions, open blockers, and
intended next step — then continues the work.
Why it exists
Agent memory is well-funded, but it standardizes agent-generated memory. Nothing standardizes the human's live working context crossing into an agent — the 15 tabs, the half-formed decision, the thing you were about to do next. That handoff is the protocol. See RFC.md for the token design (EdDSA-signed, scope-filtered, TTL-capped, revocable, audited).
We're proposing Resume Tokens as an open standard — issues and PRs welcome. The goal is for every agent framework to accept a Resume Token, the way every framework now speaks MCP.
Get a live key
Free for 10,000 hydrations/month at unihodl.app/developers.
Verify your key against the live API: https://www.unihodl.app/.well-known/jwks.json.
License
MIT — see LICENSE.
Available Tools
2 toolslist_sessionsARead-onlyIdempotent
List sessions in the workspace the configured API key can read, newest first, with cursor pagination. Read-only, no side effects. Returns {sessions: [{session_id, title, summary, captured_at, ai_tags}], next_cursor}. Pass any session_id to the resume tool for the full context; pass next_cursor back as cursor to fetch the next page (null means no more pages). Use this to discover what the human was working on when you do not already have a session_id — if you have one, call resume directly. Sandbox keys see the demo workspace. Errors: a malformed cursor or since value returns an error stating the expected format.
Inputs: limit, cursor, since. Returns: sessions, next_cursor
resumeARead-onlyIdempotent
Fetch a UNIHODL session as Resume Context — the human's open tabs, scroll positions, video timestamps, AI-tagged decision thread, partial conclusions, and intended next step. Read-only and idempotent: it never modifies the session. Use it when you have a session_id (from list_sessions or the user) and need the human's working context before continuing their task; to discover sessions instead, use list_sessions. Returns a prompt-ready text block by default, or the raw Resume Context object with format 'json'. Errors: a malformed session_id is rejected before any network call; an unknown, expired, or revoked session returns an error message stating the reason.
Inputs: session_id (required), format.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 2 tools
Each tool has a clear, distinct purpose: list_sessions discovers available sessions, while resume fetches details of a specific session. There is no overlap or ambiguity between them.
Both tools follow a verb_noun pattern (list_sessions, resume), though 'resume' is a single verb without an explicit noun. The naming is mostly consistent and readable.
With only 2 tools, the server is on the lower end of reasonable scope. It covers the basic read operations for sessions but feels minimal; additional tools like searching or filtering might be expected.
The server provides essential read functionality (listing and resuming sessions) but lacks any mutation operations or advanced filtering, leaving notable gaps for full session management.
Maintenance
Related MCP Connectors
Let AI agents use any website you're logged in to. Sessions stay encrypted on your machine.
Browser MCP for logged-in tasks. Uses your Chrome — credentials stay local. Zero-token replay.
Capture coding-agent sessions and resume them in any other agent: Claude Code, Codex, Cursor, more.
Let AI agents use your real, signed-in Chrome: tabs, clicks, forms, screenshots and video.
Related MCP Servers
AlicenseNot gradedqualityCmaintenanceToken-efficient MCP control for the Chrome session you already use—tabs, cookies, logins, and all.MIT- AlicenseNot gradedqualityBmaintenanceEnables AI agents to drive a live, persistent self-hosted Chromium session over MCP, running browser actions via Playwright/CDP, capturing page snapshots, screenshots and videos, and managing uploads and downloads in the same tabs and signed-in accounts a human user can take over at any time.Apache 2.0
- AlicenseAqualityCmaintenanceTurns your everyday Chrome into an MCP server so AI agents can operate real tabs with your logins, watch and transcribe video, read all DevTools data, and call W3C WebMCP tools published by websites. Agents get native browsing control, performance and network inspection, and page-published tools as typed MCP calls, all under your control with per-session stop and resume.56Apache 2.0
- AlicenseNot gradedqualityBmaintenanceEnables any MCP-capable AI agent to operate the user's own everyday Chrome browser with all existing login sessions intact, offering 22 tools for reading page/network data, filling forms, clicking, batch multi-step actions, and handing CAPTCHAs or confirmations back to the human — without API keys, debug ports, or launch flags.MIT