proxylens-mitmproxy 0.4.5

mitmproxy addon client for ProxyLens Server

ProxyLens for mitmproxy

The PyPI package is proxylens-mitmproxy and the Python import package is proxylens_mitmproxy. It captures HTTP and WebSocket traffic as normalized ProxyLens Server events.

It does three main things:

• propagates X-ProxyLens-HopChain
• generates a fresh X-ProxyLens-RequestId per observed hop
• synchronizes standard W3C, B3, or Jaeger trace headers, and synthesizes W3C context for fresh 32-hex traces when no standard context arrived inbound
• submits request, response, body, trailer, websocket, and error events to a server client

It can also optionally limit how many requests are allowed through the proxy at the same time, which is useful when you want more deterministic capture order from applications that cannot propagate ProxyLens headers themselves.

Setup

This subproject is self-contained under mitmproxy_addon/.

cd mitmproxy_addon
uv sync --dev

Run tests with:

uv run pytest

Basic Addon Usage

ProxyLens() now defaults to a real HTTP client for ProxyLens Server.

Current built-in clients:

• RecordingProxyLensServerClient: test/dummy client that records uploads and events in memory
• ProxyLensServerClient: real HTTP client for the server write API

Minimal example:

from proxylens_mitmproxy import ProxyLens

addon = ProxyLens(
    node_name="proxy-a",
    max_concurrent_requests_per_host=1,
)

addons = [addon]

If you do not pass node_name, the addon reads it from PROXYLENS_NODE_NAME.

export PROXYLENS_NODE_NAME=proxy-a

If you do not inject a client, the addon uses ProxyLensServerClient, which resolves the server base URL from PROXYLENS_SERVER_BASE_URL and otherwise defaults to http://127.0.0.1:8000.

export PROXYLENS_SERVER_BASE_URL=http://127.0.0.1:8000

You can still inject a custom client explicitly:

from proxylens_mitmproxy import ProxyLens, ProxyLensServerClient

addon = ProxyLens(
    client=ProxyLensServerClient(base_url="http://127.0.0.1:8000"),
    node_name="proxy-a",
)

max_concurrent_requests_per_host=None keeps the default unlimited behavior. Set it to 1 to force strict per-host serialization, or to a larger integer to allow a small bounded number of in-flight requests per host. WebSocket upgrade requests do not count toward this limit.

If you do not pass max_concurrent_requests_per_host, the addon reads it from PROXYLENS_MAX_CONCURRENT_REQUESTS_PER_HOST when set.

export PROXYLENS_MAX_CONCURRENT_REQUESTS_PER_HOST=1

Running In mitmproxy

Create a small loader script, for example run_proxylens_mitmproxy.py:

from proxylens_mitmproxy import ProxyLens

addons = [
    ProxyLens(
        node_name="proxy-a",
    )
]

Then run mitmproxy or mitmdump with that script:

cd mitmproxy_addon
uv run mitmdump -s run_proxylens_mitmproxy.py

Behavior

For each request seen by the current mitmproxy process, the addon:

1. reads any inbound X-ProxyLens-HopChain
2. appends the current node name or starts a new trace using a trace id extracted from traceparent, B3, or Jaeger headers when available, otherwise generating a new trace id
3. replaces any inbound X-ProxyLens-RequestId with a fresh ULID
4. preserves the inbound standard propagation format when W3C, B3, or Jaeger context is already present
5. synthesizes W3C traceparent and tracestate when no standard trace context arrived inbound and the resolved trace id is 32 hex, so downstream OpenTelemetry instrumentation can join the same trace automatically
6. emits normalized capture events in request-local event_index order
7. uploads binary body chunks and binary websocket payloads before emitting referencing events

The addon currently captures:

• request metadata
• request body chunks
• request trailers when exposed by mitmproxy
• request completion
• response metadata
• response body chunks
• response trailers when exposed by mitmproxy
• response completion
• websocket start, message, and end events
• request error events

Dependency Injection

ProxyLens accepts injectable collaborators for deterministic tests and custom runtime behavior:

ProxyLens(
    client=...,
    node_name="proxy-a",
    server_base_url=...,
    trace_id_generator=...,
    request_id_generator=...,
    blob_id_generator=...,
    flow_filter=...,
    max_concurrent_requests_per_host=...,
)

flow_filter(flow) can be used to skip capture for selected flows. server_base_url is only used when client is not injected. If it is unset and PROXYLENS_SERVER_BASE_URL is also unset, the addon disables itself and becomes an early no-op. max_concurrent_requests_per_host limits how many flows can be active at once for each destination host; excess flows for that host are queued inside mitmproxy until a slot becomes available. WebSocket upgrade flows are excluded from that accounting.

In-Process Test Harness

The package includes a mitmproxy-native test harness for fast integration tests:

from mitmproxy import http
from proxylens_mitmproxy import (
    ProxyLens,
    RecordingProxyLensServerClient,
    TestMitmProxy,
)


def handler(flow: http.HTTPFlow) -> None:
    flow.response = http.Response.make(200, b"ok")


client = RecordingProxyLensServerClient()
addon = ProxyLens(client=client, node_name="proxy-a")

with TestMitmProxy(proxy_lens=addon, handler=handler) as proxy:
    flow = proxy.request("GET", "https://example.test/")

assert flow.response is not None
assert client.events[0]["type"] == "http_request_started"

TestMitmProxy provides:

• request(...) / send(...) for sync tests
• arequest(...) / asend(...) for async tests
• real mitmproxy.http.HTTPFlow objects
• responder exception surfacing instead of silent logging

Package Layout

mitmproxy_addon/
├── pyproject.toml
├── README.md
├── src/proxylens_mitmproxy/
└── tests/