diff --git a/manual.pdf b/manual.pdf
index f7ad06a..8a4192d 100644
Binary files a/manual.pdf and b/manual.pdf differ
diff --git a/manual.typ b/manual.typ
index e052c0a..c1668d3 100644
--- a/manual.typ
+++ b/manual.typ
@@ -13,6 +13,12 @@
 )
 #tidy.show-module(sha-doc)
 
+#let md-doc = mod(
+  read("src/md.typ"),
+  name: "md"
+)
+#tidy.show-module(md-doc)
+
 #let misc-doc = mod(
   read("src/misc.typ"),
   name: "misc"
diff --git a/src/lib.typ b/src/lib.typ
index 2a16f0b..53a1165 100644
--- a/src/lib.typ
+++ b/src/lib.typ
@@ -1,4 +1,5 @@
 #import "misc.typ": *
-#import "sha.typ"
+#import "sha.typ": *
+#import "md.typ": *
 #import "utils.typ": *
 #import "base.typ": *
\ No newline at end of file
diff --git a/src/md.typ b/src/md.typ
new file mode 100644
index 0000000..f1ffb48
--- /dev/null
+++ b/src/md.typ
@@ -0,0 +1,259 @@
+#import "utils.typ": *
+
+#let rots = (
+  7, 12, 17, 22,  7, 12, 17, 22,  7, 12, 17, 22,  7, 12, 17, 22,
+  5,  9, 14, 20,  5,  9, 14, 20,  5,  9, 14, 20,  5,  9, 14, 20,
+  4, 11, 16, 23,  4, 11, 16, 23,  4, 11, 16, 23,  4, 11, 16, 23,
+  6, 10, 15, 21,  6, 10, 15, 21,  6, 10, 15, 21,  6, 10, 15, 21
+)
+
+#let _md5-const(i) = {
+  return int(
+    calc.floor(
+      calc.abs(
+        calc.sin(i + 1) * max-32
+      )
+    )
+  )
+}
+
+#let md5-default-iv = (
+  0x67452301,
+  0xEFCDAB89,
+  0x98BADCFE,
+  0x10325476
+)
+#let md4-default-iv = md5-default-iv
+
+/// Message Digest 5
+/// #test(
+///   `jumble.bytes-to-hex(jumble.md5("Hello World!")) == "ed076287532e86365e841e92bfc50d8c"`
+/// )
+/// ```example
+/// #bytes-to-hex(md5("Hello World!"))
+/// ```
+/// -> bytes
+#let md5(
+  message,
+  iv: md5-default-iv
+) = {
+  let message = if type(message) == str {bytes(message)} else {message}
+  assert(type(message) == bytes, message: "message must be a string or bytes, but is " + repr(type(message)))
+
+  // Complete message to multiple of 512 bits
+  let bin-str = ""
+  for char in message {
+    bin-str += z-fill(str(char, base: 2), 8)
+  }
+  let l = bin-str.len()
+  bin-str += "1"
+  let padding = calc.rem-euclid(448 - bin-str.len(), 512)
+  if padding != 0 {
+    bin-str += "0" * padding
+  }
+  let len-bin = z-fill(str(l, base: 2), 64)
+  // Reverse order to little-endian
+  len-bin = len-bin.clusters()
+                   .chunks(8)
+                   .rev()
+                   .flatten()
+                   .join()
+  bin-str += len-bin
+  let bin = bin-str.clusters().map(int)
+
+  // Split into blocks of 16 32-bit words
+  let words = bin.chunks(32).map(bin-to-int).map(switch-endianness)
+  let blocks = words.chunks(16)
+  
+  let vec = iv
+
+  for block in blocks {
+    let (A, B, C, D) = vec
+    for i in range(64) {
+      let (f, g) = if i < 16 {(
+        B.bit-and(C).bit-or(B.bit-not().bit-and(mask-32).bit-and(D)),
+        i
+      )} else if i < 32 {(
+        D.bit-and(B).bit-or(D.bit-not().bit-and(mask-32).bit-and(C)),
+        calc.rem(5 * i + 1, 16)
+      )} else if i < 48 {(
+        B.bit-xor(C).bit-xor(D),
+        calc.rem(3 * i + 5, 16)
+      )} else {(
+        C.bit-xor(B.bit-or(D.bit-not().bit-and(mask-32))),
+        calc.rem(7 * i, 16)
+      )}
+      let temp = circular-shift(
+        calc.rem(A + f + _md5-const(i) + block.at(g), max-32),
+        n: rots.at(i)
+      ) + B
+      (A, B, C, D) = (
+        D,
+        calc.rem(temp, max-32),
+        B,
+        C
+      )
+    }
+
+    vec = vec.zip((A, B, C, D)).map(p => {
+      calc.rem(p.sum(), max-32)
+    })
+  }
+
+  let digest-bytes = ()
+  for n in vec {
+    digest-bytes += (
+      n.bit-and(0xff),
+      n.bit-rshift(8).bit-and(0xff),
+      n.bit-rshift(16).bit-and(0xff),
+      n.bit-rshift(24)
+    )
+  }
+
+  return bytes(digest-bytes)
+}
+
+#let _md4-round1-shift = (3, 7, 11, 19)
+#let _md4-round2-shift = (3, 5, 9, 13)
+#let _md4-round3-shift = (3, 9, 11, 15)
+
+#let _md4-f(x, y, z) = {
+  return x.bit-and(y).bit-or(x.bit-not().bit-and(mask-32).bit-and(z))
+}
+
+#let _md4-g(x, y, z) = {
+  let xy = x.bit-and(y)
+  let xz = x.bit-and(z)
+  let yz = y.bit-and(z)
+  return xy.bit-or(xz).bit-or(yz)
+}
+
+#let _md4-h(x, y, z) = {
+  return x.bit-xor(y).bit-xor(z)
+}
+
+#let _md4-round1(a, b, c, d, x, s) = {
+  let temp = a + _md4-f(b, c, d) + x
+  return circular-shift(
+    temp.bit-and(mask-32),
+    n: s
+  )
+}
+
+#let _md4-round2(a, b, c, d, x, s) = {
+  let temp = a + _md4-g(b, c, d) + x + 0x5a827999
+  return circular-shift(
+    temp.bit-and(mask-32),
+    n: s
+  )
+}
+
+#let _md4-round3(a, b, c, d, x, s) = {
+  let temp = a + _md4-h(b, c, d) + x + 0x6ed9eba1
+  return circular-shift(
+    temp.bit-and(mask-32),
+    n: s
+  )
+}
+
+/// Message Digest 4
+/// #test(
+///   `jumble.bytes-to-hex(jumble.md4("Hello World!")) == "b2a5cc34fc21a764ae2fad94d56fadf6"`
+/// )
+/// ```example
+/// #bytes-to-hex(md4("Hello World!"))
+/// ```
+/// -> bytes
+#let md4(
+  /// -> str | bytes
+  message,
+  /// -> array
+  iv: md4-default-iv
+) = {
+  let message = if type(message) == str {bytes(message)} else {message}
+  assert(type(message) == bytes, message: "message must be a string or bytes, but is " + repr(type(message)))
+
+  // Complete message to multiple of 512 bits
+  let bin-str = ""
+  for char in message {
+    bin-str += z-fill(str(char, base: 2), 8)
+  }
+  let l = bin-str.len()
+  bin-str += "1"
+  let padding = calc.rem-euclid(448 - bin-str.len(), 512)
+  if padding != 0 {
+    bin-str += "0" * padding
+  }
+  let len-bin = z-fill(str(l, base: 2), 64)
+  // Reverse order to little-endian
+  len-bin = len-bin.clusters()
+                   .chunks(8)
+                   .rev()
+                   .flatten()
+                   .join()
+  bin-str += len-bin
+  let bin = bin-str.clusters().map(int)
+
+  // Split into blocks of 16 32-bit words
+  let words = bin.chunks(32).map(bin-to-int).map(switch-endianness)
+  let blocks = words.chunks(16)
+  
+  let vec = iv
+
+  for block in blocks {
+    let vec2 = vec
+    
+    for i in range(16) {
+      vec2.at(0) = _md4-round1(
+        ..vec2,
+        block.at(i),
+        _md4-round1-shift.at(calc.rem(i, 4))
+      )
+      vec2.insert(0, vec2.pop())
+    }
+
+    for i in range(16) {
+      let j = calc.rem(i * 4, 16) + calc.div-euclid(i, 4)
+      vec2.at(0) = _md4-round2(
+        ..vec2,
+        block.at(j),
+        _md4-round2-shift.at(calc.rem(i, 4))
+      )
+      vec2.insert(0, vec2.pop())
+    }
+
+    for i in range(16) {
+      let j = bin-to-int(
+        z-fill(
+          str(i, base: 2),
+          4
+        ).clusters()
+         .map(int)
+         .rev()
+      )
+
+      vec2.at(0) = _md4-round3(
+        ..vec2,
+        block.at(j),
+        _md4-round3-shift.at(calc.rem(i, 4))
+      )
+      vec2.insert(0, vec2.pop())
+    }
+
+    vec = vec.zip(vec2).map(p => {
+      calc.rem(p.sum(), max-32)
+    })
+  }
+
+  let digest-bytes = ()
+  for n in vec {
+    digest-bytes += (
+      n.bit-and(0xff),
+      n.bit-rshift(8).bit-and(0xff),
+      n.bit-rshift(16).bit-and(0xff),
+      n.bit-rshift(24),
+    )
+  }
+
+  return bytes(digest-bytes)
+}
\ No newline at end of file
diff --git a/src/sha.typ b/src/sha.typ
index ec09ff4..dd2b9f0 100644
--- a/src/sha.typ
+++ b/src/sha.typ
@@ -42,15 +42,18 @@
 /// -> bytes
 #let sha1(
   /// Message to hash
-  /// -> str
+  /// -> str | bytes
   message,
   /// Initial vector
   /// -> array
   iv: sha1-default-iv
 ) = {
+  let message = if type(message) == str {bytes(message)} else {message}
+  assert(type(message) == bytes, message: "message must be a string or bytes, but is " + repr(type(message)))
+
   // Complete message to multiple of 512 bits
   let bin-str = ""
-  for char in bytes(message) {
+  for char in message {
     bin-str += z-fill(str(char, base: 2), 8)
   }
   let l = bin-str.len()
diff --git a/src/utils.typ b/src/utils.typ
index 3b7cd2f..0f05eda 100644
--- a/src/utils.typ
+++ b/src/utils.typ
@@ -97,4 +97,23 @@
   }
   let high-bits = x.bit-rshift(32 - n)
   return x.bit-lshift(n).bit-or(high-bits).bit-and(mask-32)
+}
+
+/// Switches the endianness of the given value (32-bit integer)
+/// -> number
+#let switch-endianness(
+  /// -> number
+  value
+) = {
+  let a = value.bit-rshift(24)
+  let b = value.bit-rshift(16).bit-and(0xff)
+  let c = value.bit-rshift(8).bit-and(0xff)
+  let d = value.bit-and(0xff)
+
+  let a2 = d.bit-lshift(24)
+  let b2 = c.bit-lshift(16)
+  let c2 = b.bit-lshift(8)
+  let d2 = a
+
+  return a2.bit-or(b2).bit-or(c2).bit-or(d2)
 }
\ No newline at end of file