BugTape integrates production error reporting with coding agents via the Model Context Protocol, allowing tools like Claude Code or Cursor to analyze logs, stack traces, and browser replays. While the platform automates evidence collection, developers must configure runtime-specific SDKs and manage granular permissions to control agent access. Crucially, the system does not capture request or response bodies, meaning agents may lack the full context needed to diagnose complex validation failures. Teams should treat agent-generated pull requests as proposals rather than verified fixes, requiring independent regression testing to ensure production stability.
BugTape lets authorized coding agents read production-error reports over the Model Context Protocol (MCP). Its browser SDK can collect replay, failed requests, console output and stack traces into issues that an agent investigates before proposing a pull request. Developers still decide what ships, and the available evidence depends on the runtime and capture settings, according to BugTape’s product page.
The public-launch date comes from its October 28, 2026, Show HN submission; the product page has no formal announcement timestamp. Some capabilities predate that introduction: BugTape’s changelog lists MCP access on every plan on October 1 and OAuth 2.1 sign-in on October 2.
The practical question is what reaches the agent, under whose authorization and at what cost. The analysis below uses BugTape’s published documentation, without independent testing.
Setup Starts With One Command, Then Capture Configuration
Run this command inside your application’s repository, using Node.js 20 or newer:
npx -y bugtape init
BugTape says the command creates or selects a project and provides installation steps. You sign in and approve access in a browser, then add the appropriate SDK, choose privacy settings and send a test report. The command starts setup; it does not eliminate application integration. See the product’s setup instructions.
Connecting an agent is a separate step. Choose the command for your client:
The CLI prints a link and code. Open the link, check the code and approve access. The browser can be on another machine, which makes this flow useful for remote agents without a local browser, according to the MCP installation guide.
The remote MCP endpoint is https://app.bugtape.ai/mcp. BugTape advertises compatibility with Claude Code, Cursor, Codex, ChatGPT, Grok and HTTP MCP clients. Configuration details differ:
Claude Code: Supports URL-based OAuth setup, followed by sign-in through /mcp.
Cursor: The connection command writes .cursor/mcp.json; selects user-wide settings.
Codex: For bearer-token setup, the guide recommends an environment variable such as BUGTAPE_MCP_TOKEN. Its configuration stores the variable’s name, keeping the token out of the repository.
These distinctions matter when sharing project configuration. A server URL can belong in configuration; a personal access token should stay out of source control and shell history.
Browser Replay Does Not Extend to Every Runtime
BugTape’s supported platforms have different capture capabilities. Its platform summary lists:
Runtime
Installation
Documented capture
Web
npm i @bugtape/sdk
Browser replay, requests, console and stack
Node.js
npm i @bugtape/node
Server errors with route context
Python
pip install bugtape
Exceptions with context
Swift, alpha
Swift package
Handled errors; no fatal-crash capture or native replay
Browser replay is reconstructed page activity from DOM events, not a screen recording. BugTape keeps a short browser buffer and sends it with a supported error or user report. It says it does not record every visit, and replay can be unavailable.
Server reports have a different shape. BugTape’s production-errors guide describes server evidence as the error, stack, route pattern, service and release, without browser replay or console capture.
Request and response bodies are not captured. That is a consequential boundary: an agent may see a failed request’s method, URL, status and timing while lacking the payload that triggered it. A validation failure caused by one unusual field could therefore require a separate, sanitized reproduction.
My view is that this is the most important qualification to the launch pitch. The agent receives a useful evidence bundle, but its completeness varies. Teams should inspect a test report from their own stack before deciding which investigations they can delegate.
Capture Keys and Agent Access Are Separate
BugTape separates sending reports from reading them. Capture keys beginning with bt_live_ or bt_test_ cannot authenticate to MCP. An agent needs an approved connection or personal access token, with access limited to selected projects and permissions, according to the installation guide.
The permission choices also affect what an agent can do:
Pick up and fix grants subscribe, get_repro_context and suggest_fix.
Follow and read omits suggest_fix.
Sign-in includes create_bug unless it is unticked.
setup_capture is added when explicitly selected.
These are permission names, not necessarily callable tool names. BugTape specifically warns that there is no subscribe tool.
I’d begin with narrow project access and the follow-and-read option, then expand permissions once the team has checked what reports contain. That keeps the first evaluation focused on evidence quality before adding proposed changes or capture setup.
Input Masking Still Leaves a Page-Text Decision
Inputs are masked by default, but visible page text can still be recorded. Strict privacy mode masks that text too, according to the product page.
That distinction matters for dashboards, support tools and account pages. Sensitive information often appears in rendered text after a form has been submitted. Default input masking alone would not cover that case.
The September 30 changelog entry also describes sealed client details: text marked with data-bugtape-token leaves the browser as a placeholder, sealed using a public key. BugTape says it stores the sealed content without being able to read it; a team uses its private key to open it in the console.
That is a specific browser-text feature. It should not be interpreted as a blanket encryption claim covering every report field or server error.
Free Includes 150 Credits, Not Unlimited Agent Reads
BugTape’s published pricing separates issue limits from credits. Its MCP guide explains the charge for reading captured events.
Plan or purchase
Published price
Relevant limits and MCP costs
Free
$0
One project, three members, 100 new issues and 150 credits monthly; seven-day event retention
Prepaid credits
$10 for 2,000 credits
Two credits per agent read of a bug’s captured events
Pro
$29 per month
Unlimited projects; agent reads over MCP included
Free also charges two credits per agent read of captured events. If all 150 monthly credits went to those reads, the allowance would cover 75 reads. That is arithmetic from BugTape’s published rate, not a vendor usage estimate or a promise of 75 completed fixes. Repeated reads can concern the same issue.
This makes agent behavior relevant to cost. Pulling a ranked inbox and investigating selected issues is a more controlled starting point than repeatedly fetching every report.
There is also a workflow distinction. BugTape’s production-errors guide says live waiting through listen requires Pro or prepaid credits. Free users can send an issue to an agent, then ask it to retrieve the ranked list. MCP access on Free therefore does not mean every background handoff mode is included.
A Linked Pull Request Is a Proposal, Not Verification
BugTape states the central limitation plainly:
“An agent response or linked pull request does not prove the fix works.”
That warning appears on the product page. A stack trace identifies where execution failed; it may not identify the underlying cause. Replay can show the sequence before a failure while missing server state or an uncaptured request body.
A sensible review process is to reproduce the failure where possible, check the proposed change against the affected release and add a regression test. The agent’s explanation and linked evidence help reviewers navigate the problem; they do not substitute for those checks.
BugTape also documents release-based regression monitoring. After a fix is approved, it can watch for the failure in later releases, including commit-SHA or build-ID releases. Its changelog notes that reports without a release receive a warning.
Release identifiers are therefore worth configuring during setup. They connect observed failures to deployed code and make later recurrence checks more useful. Even then, the absence of another captured error is weaker evidence than a test demonstrating that the original failure now succeeds.
Frequently Asked Questions
3 questions
1
Does a User Have to Report the Bug?
No. The browser SDK can automatically report supported errors and failed requests, subject to capture settings and rate limits. Users can also submit reports, according to BugTape.
2
What Does the Free Tier Include?
One project, three members, 100 new issues and 150 credits per month, plus seven-day event retention. No credit card is required. Captured-event reads cost two credits each on Free.
3
Can I Try BugTape Without an Account?
Yes. npx -y bugtape --help requires no sign-in. BugTape also offers a local browser demo and a no-sign-in demo MCP endpoint at https://app.bugtape.ai/demo/mcp. Access to your own workspace requires authorization.
