wechat-sendmsg-cgi-analysis.md 3.52 KB

WeChat SendMsg CGI Analysis

Date: 2026-06-30 Source: smali decompilation of WeChat 8.0.74 (smali_classes12/w11/r0.smali)

CGI Endpoint

POST /cgi-bin/micromsg-bin/newsendmsg
FuncID: 522 (0x20a)
MMFunc: 237 (0xed)

Protobuf Schema

Request: r45/ez5

ez5 {
    d: int                    // message count?
    e: LinkedList<ll4>        // list of messages
}

Message Element: r45/ll4

ll4 {
    d: du5 {                  // content descriptor
        d: String             // content type descriptor
        e: boolean            // flag
    }
    e: String                 // talker / content reference
    f: int                    // message type (1=text, 3=image, etc.)
    g: int                    // create time (timestamp)
    h: int                    // hash code
}

Response: r45/fz5

fz5 {
    e: LinkedList<f4>          // response items
    f: int                     // result code
}

Transport Layer

NOT direct HTTP. All CGI requests go through mars::stn:

  1. Protobuf body is ECDH-encrypted with mars session key
  2. Sent over mars QUIC/TCP long connection
  3. NetScene framework (w11/r0 extends m1) dispatches via mars

Code Reference (r0.smali)

CGI Path Assignment (lines 1580-1618)

# Build CGI request
new-instance v1, Lcom/tencent/mm/modelbase/l;
invoke-direct {v1}, Lcom/tencent/mm/modelbase/l;-><init>()V

new-instance v2, Lr45/ez5;          # Request protobuf
invoke-direct {v2}, Lr45/ez5;-><init>()V
iput-object v2, v1, Lcom/tencent/mm/modelbase/l;->a:Lcom/tencent/mm/protobuf/f;

new-instance v2, Lr45/fz5;          # Response protobuf
invoke-direct {v2}, Lr45/fz5;-><init>()V
iput-object v2, v1, Lcom/tencent/mm/modelbase/l;->b:Lcom/tencent/mm/protobuf/f;

const-string v2, "/cgi-bin/micromsg-bin/newsendmsg"
iput-object v2, v1, Lcom/tencent/mm/modelbase/l;->c:Ljava/lang/String;

const/16 v2, 0x20a                  # FuncID = 522
iput v2, v1, Lcom/tencent/mm/modelbase/l;->d:I

const/16 v2, 0xed                   # MMFunc = 237
iput v2, v1, Lcom/tencent/mm/modelbase/l;->e:I

const v2, 0x3b9acaed                # Session flag
iput v2, v1, Lcom/tencent/mm/modelbase/l;->f:I

getType() Method (line 1994)

.method public getType()I
    .locals 1
    const/16 v0, 0x20a
    return v0
.end method

onGYNetEnd Response Handler (line 2002)

# Checks response type == 0x20a
# Parses fz5 response
# e.f (result code) determines success
# Calls PluginMessengerFoundation.Di() to update message status

Why Direct HTTP Call Fails

  1. No HTTP listener: /cgi-bin/micromsg-bin/newsendmsg is NOT a REST endpoint
  2. mars session required: Body encrypted with ECDH session key
  3. Session establishment: Requires mars ECDH handshake (native code)
  4. Cookie/header auth: Session tokens bound to mars connection, not HTTP cookies

Fallback Attempt (curl from Termux)

# This WILL fail — no direct HTTP access
curl -X POST 'https://szlong.weixin.qq.com/cgi-bin/micromsg-bin/newsendmsg' \
  -H 'Content-Type: application/octet-stream' \
  --data-binary @ez5_request.bin
# → Connection refused or 404 (endpoint not HTTP-exposed)

Comparison with XHS BCIM

XHS BCIM WeChat mars
Protocol Custom Longlink Java mars C++ native
Encoding JSON Protobuf
Encryption None (plain TCP) ECDH + AES
CGI path N/A (pure socket) /cgi-bin/micromsg-bin/newsendmsg
HTTP fallback None None (CGI transported via mars)
Replayability Can replay via xhshook proxy Cannot — needs mars session