Skip to main content
Glama
swipeflow

SwipeFlow MCP Server

by swipeflow
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Remote

SwipeFlow MCP Server

Human-in-the-loop approvals for AI agents, over the Model Context Protocol.

An agent asks a person to review something, gets a durable handle back, and resumes once a decision has been made. Human review takes minutes or days, so the wait is asynchronous by design — the agent is never required to hold a connection open.

Status: early development. The tool surface is being built against MCP 2026-07-28. Not yet ready for production use.

Why this exists

Agents increasingly produce work that a human should approve before it ships — content, code changes, outbound messages, spending decisions. The awkward part is never the approval itself, it's the waiting: humans are slow and agents are not built to block.

This server makes that wait a first-class protocol concern.

Related MCP server: QMCP

Design

Three properties shape everything else:

Stateless. The server holds no task store, no session store, no database. Task state lives in SwipeFlow, because a review is a task. That is what gives resumability across restarts, horizontal scaling with no coordination, and self-hosting with no infrastructure.

Decisions are outcomes, not failures. Approved, rejected, and changes-requested all complete normally and carry an outcome. Failure is reserved for genuine errors, so an agent can branch on a human's judgement without treating disagreement as a fault.

Works with or without the Tasks extension. Support varies across MCP clients, so every asynchronous tool has a synchronous fallback with identical semantics and identical IDs. Tasks is an upgrade in how a handle is delivered, never in what it means.

Development

Requires Node.js 20 or newer.

npm install
npm run verify     # boundary checks, lint, typecheck, tests
npm run build

Individual steps:

npm run check:deps      # dependency allowlist
npm run check:boundary  # copied-code fingerprints
npm run lint
npm run typecheck
npm test

The boundary checks

This repository is public; the SwipeFlow backend is not. The MCP server reaches SwipeFlow only through its public REST API — never a database, never a shared private module.

Two guards enforce that mechanically rather than by review discipline:

  • check:deps — every dependency must be declared in allowed-dependencies.json. Adding one is a deliberate act. Stale entries also fail, so the allowlist cannot quietly accumulate grants.

  • check:boundary — scans source for the fingerprints of backend code having been copied across: database drivers, backend-only credential surfaces, relative imports that escape the repository.

Both run in CI ahead of lint and tests, and both are themselves covered by tests in test/boundary.test.ts — a guard nobody exercises is a guard that silently stops working.

Contributing

Issues and discussion: https://github.com/swipeflow/swipeflow-mcp/issues

License

Apache 2.0

Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessSyncing

Resources

Related MCP Connectors

Related MCP Servers

  • Interactive Feedback MCP

    ISimon3
    F
    license
    B
    quality
    D
    maintenance
    MCP server that enables human-in-the-loop workflow in AI-assisted development tools by allowing users to provide direct feedback to AI agents without consuming additional premium requests.
    1
    1
    -

  • QMCP

    quaternionmedia
    F
    license
    Not graded
    quality
    B
    maintenance
    A FastAPI-based Model Context Protocol server that enables tool discovery, invocation history, and human-in-the-loop interaction workflows. It features a Python client and CLI for managing automated tasks that require manual approval and persistence.
    -
  • Datashift MCP Server by datashift-io

    Datashift MCP Serverofficial

    datashift-io
    A
    license
    Not graded
    quality
    D
    maintenance
    Enables AI agents to submit tasks for human or AI review and receive decisions via MCP tools, adding human review checkpoints to workflows.
    MIT

  • call-a-human-mcp

    nishantmodak
    A
    license
    A
    quality
    A
    maintenance
    An MCP server that enables AI agents to pause and request human approval or information via Slack, Telegram, or macOS dialogs before proceeding with actions.
    2
    15
    Apache 2.0