Private beta MCP

Flock MCP Server

Bring product review into your coding workflow: inspect the repository, plan a test, review the evidence, and fix the flow.

Flock exposes 14 tools over Streamable HTTP. Connect with OAuth and a preregistered client, then choose read-only access or opt in to test actions.

Connection

Use the hosted setup guide for the exact Codex or Claude Code configuration and login commands. Other clients need a preregistered client ID and callback. Start with read-only access; add the test scope only when you want launch, retry, and cancel actions.

Connection requirements

Use Streamable HTTP with OAuth and PKCE.
Your client ID and callback must be preregistered. Arbitrary dynamic client registration is not supported.
Flock does not issue an API key or client secret for this connection.

Hosted server settings

server URL: https://mcp.flocksynthetics.com/mcp
transport: Streamable HTTP
authorization: OAuth with PKCE
client: preregistered client ID and callback
read scope: https://mcp.flocksynthetics.com/mcp/read
test scope (opt-in): https://mcp.flocksynthetics.com/mcp/test

Try this with your coding agent

Inspect this repository and my current changes. Identify the user flows most likely to be affected, and propose a focused Flock review with suitable personas and goals.

Use Flock to confirm my workspace, review readiness, and estimate usage. Use a reachable preview or staging URL. If I have only a localhost server, stop and explain what reachable target is needed. For a flow that needs login, use an available saved sign-in profile; do not ask me to paste credentials.

Show me the proposed target, flows, configuration, and estimate, then wait for my approval before launching. After the review, inspect the findings, available evidence images, and repair guidance. State what the evidence supports and what remains uncertain.

Propose a scoped fix. Once I approve it, make the change and run the relevant local checks. Prepare a review of the same flow against the updated preview, and get fresh approval before launching the retest. Compare the new evidence before calling the issue fixed.

The coding agent inspects your repository. Flock's configuration recommendations use your goal and the available catalog; the recommendation tool does not inspect your application.

Operator workflow

Inspect the repository and changes

Your coding agent uses its repository access to identify affected flows and propose personas and goals.

Check readiness and usage

Confirm the workspace and reachable target, select saved sign-in if needed, and review the configuration and estimate.

Approve and run

Approve the proposed review before launch. Test access and workspace permission are required.

Inspect findings and evidence

Read the complete results, inspect available images, and retrieve the relevant repair guidance. Keep unsupported observations separate from confirmed issues.

Make a scoped fix

Review the proposed change, then let the coding agent edit the relevant code and run local checks within your approval.

Review the same flow again

Check the updated target and estimate, approve the retest, and compare the new evidence before calling the issue fixed.

An estimate is not a price guarantee or a promise of a free test. Supply a unique idempotency key; if the response is interrupted, retry the exact same payload and key. Follow every findings page and prompt chunk before treating a review as complete.

Example tool calls

Replace placeholder IDs with values returned by Flock. Before starting a test, call estimate_run with the proposed configuration and review readiness and usage. After approval, call launch_run with a unique idempotency key. Use get_run_status with its batch_id before requesting final results. For a finding prompt, use the retrieval selector returned by get_review; follow content_next_cursor until the prompt is complete.

1. Recommend a test

tool: recommend_test_configuration
arguments:
  workspace_id: WORKSPACE_ID
  url: https://preview.example.com/signup
  goal: Create a shared workspace and invite a teammate

2. Retrieve run results

tool: get_run_results
arguments:
  workspace_id: WORKSPACE_ID
  run_id: RUN_ID

3. Retrieve a finding prompt

tool: get_review
arguments:
  workspace_id: WORKSPACE_ID
  review_id: REVIEW_ID
  prompt_target: finding
  finding_id: FINDING_ID

Protocol

Recommend, test, and review through tools.

Your client can recommend a test, estimate its usage, launch it after approval, and read the results. Evidence images and repair prompts are retrieved through tools; Flock does not expose them as native MCP Resources or Prompts.

The 14 Flock tools

Discover and prepare

account, workspaces, list_auth_profiles, recommend_test_configuration, estimate_run. Confirm the workspace, check saved sign-in, and estimate a proposed test. Recommendations use your goal and the available catalog; they do not inspect the application.

Run and inspect

launch_run, get_run_status, get_run_results, get_run_image, retry_run, cancel_run. Launch only after approval, follow the lifecycle, and retrieve available evidence images. Launch, retry, and cancel require the test scope and workspace permission.

Read reviews and repair prompts

list_reviews, get_latest_review, get_review. Find workspace reviews, read findings, and retrieve available repair prompts. These are tool calls, not native MCP Resources or Prompts.

Boundaries

Keep permissions and evidence status explicit.

Read access includes recommendations, estimates, status, results, review reads, and available run images. Test access adds launch, retry, and cancel, subject to workspace policy and spending controls.
For a target that requires login, complete a saved sign-in in Flock and select that profile. Never paste passwords, tokens, cookies, or browser storage into chat.
Results can be partial, not ready, or unavailable. Respect each result’s status and reason; a missing image is not evidence that no issue exists.
Read images only through get_run_image using an artifact_id returned by get_run_results. Do not construct image IDs or fetch arbitrary URLs.
Repair prompt availability and text are separate. Follow the supplied retrieval selector and ordered chunks, preserving their byte counts and hashes. An unavailable full-review prompt does not invalidate an available finding prompt.

Protocol references

Connection and protocol references.

Use the hosted Flock guide for current client setup, scopes, tool inputs, paging, and prompt retrieval. The MCP specification explains the underlying protocol.

Flock setup guide Latest specification

Want MCP access in the private beta?

Tell us which client and workflow you want to connect. We can confirm private beta access and the setup requirements for your client.

Request private beta access