Skip to main content

Planned private beta workflow

Flock MCP Server

Connect AI clients to Flock reviews through Model Context Protocol, so an operator or coding agent can request a synthetic review and inspect the evidence that came back.

MCP gives Flock a standard way to expose actions, context, and reusable prompts without inventing one-off integrations for every client.

Protocol shape

One review loop, exposed as tools, resources, and prompts.

MCP is the protocol layer between an AI application and an external capability server. For Flock, that means the client can request a review, read the resulting artifacts, and use a prompt template to turn evidence into implementation work.

Transport options

  • Streamable HTTP for a hosted Flock server managed outside the local machine.
  • OAuth-based authorization for remote flows when credentials are required.
  • No local package, environment variable, or long-lived API key should be required for the customer path.

Capabilities Flock plans to expose

Tools

Actions the client can request, such as starting a review, checking run status, or fetching a rendered artifact.

Resources

Readable context the client can inspect, such as batch digests, per-run friction logs, and screenshot evidence.

Prompts

Reusable workflow templates for turning evidence into fix briefs, regression prompts, review summaries, or follow-up questions.

Connection shape

Private beta access will enable the hosted Flock MCP endpoint for your workspace. The example below uses a reserved placeholder host until beta workspace endpoints are enabled.

Remote MCP server

{
  "mcpServers": {
    "flock": {
      "url": "https://mcp.flocksynthetics.example/mcp"
    }
  }
}

Example review loop

1. Request a review tool

tool: flock_run_ux_test
input:
  url: https://preview.example.com/signup
  personas:
    - skeptical-buyer
  goal: Create a shared workspace and invite a teammate
  environment: preview
  wait: true
  success_signals:
    - type: text_visible
      value: Invitation sent

2. Read a finding resource

resource: flock://run/run_20260526_1420/friction

severity: major
summary: Workspace setup ends without a clear next step.
evidence:
  screenshot: artifacts://run_20260526_1420/step-05.png
  observed_text: No invite, onboarding, or dashboard CTA was visible.
fix_prompt: Add the next action after workspace creation.

3. Use a fix prompt

prompt: fix-friction-event
arguments:
  batch_id: batch_20260526_1420
  event_index: 1

prompt_output:
  fix: Add the next action after workspace creation.
  preserve:
    - current workspace creation flow
    - existing invite permissions
  verify:
    - rerun the same persona and goal

Operator workflow

01

Connect from an AI client

Add the hosted Flock MCP URL to a client that supports remote servers over Streamable HTTP.

02

Approve tool calls

The client stays in control. Review-starting tools should be visible to the operator before they run against a URL.

03

Read evidence back as resources

Findings should come back with screenshot evidence, observed text, severity, and a scoped fix prompt.

04

Hand off to a human or coding agent

Use prompts to summarize a finding, draft an implementation brief, and define the rerun needed to verify the fix.

Boundaries

A standard protocol still needs explicit trust boundaries.

  • The hosted MCP endpoint is enabled during private beta access; Flock-internal stdio mode is for contributors working on the server itself.
  • Do not paste production secrets or customer credentials into an MCP client unless that workflow has been explicitly approved for your workspace.
  • Authenticated or private targets need an agreed handling path before Flock runs against them.
  • MCP is an integration surface, not a replacement for evidence review: findings still need attached artifacts.

Protocol references

Built around the current MCP language.

We use the official MCP terms for this page: clients connect to servers, servers expose tools, resources, and prompts, and remote servers use the current Streamable HTTP transport model.

MCP overview Latest specification

Want MCP access in the private beta?

Tell us which AI client and target workflow you want to connect. We will confirm the current server shape before enabling the workspace.

Request private beta access