AI agents that read the report and work the fix

One URL, nothing to host

The server is at https://backend.flawsnap.com/mcp. There is no package to install or server to run. You connect a client to that URL and sign in, or give it a personal API key.

Pick how you connect

claude mcp add --transport http flawsnap https://backend.flawsnap.com/mcp

With OAuth you pick the organization and the access level, then confirm with your password. ChatGPT availability depends on your ChatGPT plan. Full setup for each client is in the MCP docs.

A typical fix loop

1. 1

   List my tasks

   list_my_tasks returns the cards assigned to you.

2. 2

   Start one

   start_task moves the card to In progress so your team can see it is taken.

3. 3

   Read the evidence

   get_report returns the description, steps and the latest 50 console and network entries; get_report_screenshot returns the annotated image.

4. 4

   Fix it in your code

   The agent works in your repository as it normally would.

5. 5

   Finish

   finish_task moves the card to QA or Done with a summary and a link to the pull request.

What agents can and cannot do

There are 26 tools. Fourteen read: who you are, your projects, tasks, reports, screenshots, board, cards, members and docs, plus search and fetch (which ChatGPT deep research uses). Twelve write, and only on Read & write access: comment on, update and track reports; create, update, move and comment on cards; start and finish tasks; create, update and comment on docs.

Agents cannot delete anything, and cannot change settings, members, invites, share links or API keys.

Safety rails

• Read only is the default. Read & write is offered only when the app asks for it.
• Every action is recorded as you, marked "via agent" with the app or key name, so the team can see what an agent did.
• Report text is treated as untrusted. Titles, descriptions, logs and comments are wrapped as untrusted_content, and the agent is told not to follow instructions inside them. Anyone can write a report, so still review what your agent does.
• One connection is one organization. A connection lasts 90 days.
• A member limited to some projects can reach only those projects through MCP too.

Personal API keys

For coding agents, create a key on the API keys page. Keys start with snaplog_pat_ (from the product's earlier name), are shown once and stored only as a hash. Choose Read only or Read & write, and an expiry of 30 days, 90 days or a year; read-only keys can also be set never to expire. The same page lists connected apps so you can disconnect them.

Questions

Is MCP access included on Starter?

No. AI connectors for Claude and ChatGPT and API keys for coding agents are on Pro, at $19 a month per organization with unlimited members.

Can an agent close every bug on my board by mistake?

It can only act with the access you grant. Start with a read-only connection, and everything it changes is recorded as you, marked via agent, so your team can see it. It can never delete anything.

Does Flawsnap train AI models on my reports?

Flawsnap does not sell your data or use it for advertising. What a connected assistant does with tool results is governed by that assistant's own terms.

Related

• Connect AI agents (MCP)
• AI bug triage with MCP: how it works and where to be careful
• Pricing
• A Kanban board where bug reports become work
• Bug reporting for SaaS product teams