Skip to main content
MCP & API

Your debug data, for your AI assistant and your scripts

Buggregator already holds what your app did: the exception, the dump, the profile, the email. The built-in MCP server lets your AI assistant read it, so it can explain a failure instead of guessing. The REST API lets your tests and scripts use the same data — wait for an email, read a code from it, set up mock answers.

Get started MCP is off by default · one switch turns it on
127.0.0.1:8000

The assistant lists recent errors, opens one, and reads the stack trace — the same data you see in the UI. Example data.

What your assistant and your scripts can do

19 MCP tools, 31 with the Xdebug proxy. The REST API covers the rest.

🐞

Read the error, not a guess

Your assistant lists recent events and opens a Sentry error in a compact form: message, exception chain, stack frames with file and line, environment and tags.

🔥

Find what’s slow

From an XHProf profile it gets the summary, the top functions by time or memory, and the call graph — and names the hot function and its callers.

📦

Dumps as plain text

Variable dumps come back as plain text, without HTML, so the assistant reads values, not markup.

🎯

At your breakpoint

With the Xdebug proxy on, the assistant reads the paused session — line, stack, variables. With no IDE attached, it can also step, continue, stop and set breakpoints, and you see each step on the Xdebug page. That's twelve tools: five read, one lists breakpoints, six control the session. xdebug.allow_control: false keeps them to reading. Evaluating PHP (xdebug_eval) also needs xdebug.allow_eval: true.

✉️

Emails in your tests

REST API: wait until an email arrives, search by recipient or subject, and pull out its links or one-time codes. Built for E2E sign-up and password-reset flows.

🧪

Mock APIs, from your assistant or a script

Your assistant can create, edit and delete mock rules, preview what a rule would answer, and turn a captured request into a rule. Give it an OpenAPI spec, and it makes one rule per endpoint, with example or fake data. Scripts can manage rules and import specs through the REST API.

Setup in 60 seconds

No registration. No account. Start it and point your app at it.

1

Turn MCP on

Requires Buggregator v2.6.0 or later; the Xdebug step and breakpoint tools need v2.7.0. MCP is off by default. Turn it on with one setting. Locally it listens on a Unix socket; in Docker, use HTTP on port 8001.

MCP_ENABLED=true

MCP has its own access rules, separate from SSO. Over HTTP it has no token unless you set MCP_AUTH_TOKEN — keep the port on 127.0.0.1, as above. Six of the mock tools change your mock rules, and they are on by default. Turn them off with mcp.mock_write: false (or MCP_MOCK_WRITE=false).

2

Connect your assistant

On the same machine, Claude Code starts buggregator mcp, which connects to the running server. With Docker, connect over HTTP at http://localhost:8001/mcp.

$ claude mcp add buggregator -- buggregator mcp
3

Ask, or call the API

Ask in plain words — the assistant picks the tools. Or call the REST API from a test or a script. With SSO on, the API needs the token from your browser sign-in.

"Why did checkout fail?"
  → events_list (type: sentry) → sentry_event

"I profiled this endpoint. Why is it slow?"
  → events_list (type: profiler) → profiler_summary → profiler_top (metric: excl_wt) → profiler_call_graph

"Why is $total zero here?"   (stopped at a breakpoint)
  → xdebug_location → xdebug_variables → xdebug_property ($cart->items)

"Set a breakpoint where the order total is computed, run the checkout, and tell me why it's negative."
  → xdebug_breakpoint_set → runs the request (curl) → xdebug_wait → xdebug_variables

"Mock the payments API from ./openapi.yaml."
  → mock_import_openapi → mock_rules_list

Works with your AI tools and your test stack

MCP for assistants, plain HTTP for everything else.

Claude Code— local via buggregator mcp, or HTTP
Codex— HTTP config in ~/.codex/config.toml
Cursor— the same command, or the HTTP URL
Any MCP client— Unix socket via buggregator mcp, or HTTP on port 8001
E2E tests— any framework that can call HTTP: wait for an email, read its code
Scripts and CI— curl or any HTTP client against /api/

Ready to try it?

One command to start. No account needed.

Also built into Buggregator