diff --git a/README.md b/README.md
index cd15c62..accccb9 100644
--- a/README.md
+++ b/README.md
@@ -77,3 +77,137 @@ A HCL program is composed of expressions.
 	function body
 )
 ```
+
+
+## HCL Exchange Protocol
+
+The HCL library contains a simple server/client libraries that can exchange
+HCL scripts and results over network. The following describes the protocol
+briefly.
+
+### Request message
+TODO: fill here
+
+### Reponse message
+
+There are two types of response messages.
+ - Short-form response
+ - Long-form response
+
+A short-form response is useful when you reply with a single unit of data.
+A long-form response is useful when the actual data to return is more complex.
+
+#### Short-form response
+A short-form response is composed of a status line. The status line may span 
+across multiple line if the single response data item can span across multiple
+lines without ambiguity. A short-form response begins with a status word. 
+The status word gets followed by an single data item.
+
+There are 2 status word defined.
+ - .OK
+ - .ERROR
+
+The data must begin on the same line as the status word and there should be 
+as least 1 whitespace characters between the status word and the beginning of
+the data. The optional data must be processible as a single unit of data. The
+followings are accepted:
+
+ * unquoted text line
+   ** The end of the data is denoted by a newline character. The newline
+      character also terminates the status line.
+   ** Leading and trailing spaces must be trimmed off
+ * quoted text
+   ** If the first meaningful character of the option data is a double quote,
+      the option data ends when another ordinary double quote is encounted.
+   ** If a double quote is preceded by a backslash(\"), the double quote becomes
+      part of the data and doesn't end the data.
+   ** Not only the double quote, any character character escaped by a preceding
+      backslash is treated literally. (e.g. \\ -> a single back slash, \n -> n)
+   ** Trailing spaces after the ending quote must be ignored until a newline
+      character is encounted. The newline character terminates the status line.
+    
+Take note of the followings when parsing a short-form response message
+ * Whitespace characters before the status word shall get ignored.
+
+See the following samples.
+
+  .OK authentication success
+
+  .ERROR double login attempt
+
+  .OK "authentication\twas\tsuccessful"
+
+  .OK "this is a multi-line
+    string message"
+
+
+#### Long-form response
+
+A long-form response begins with the status word line. The status line 
+should be composed of the status word and a new line. The status line must
+get followed by data format line and the actual response data. Optional 
+attribute lines may get inserted between the status line and the data format
+line.
+
+The data format line begins with .DATA and it can get followed by a data length
+or the word 'chunked'.
+
+ * .DATA <NNN>
+ * .DATA chunked
+ 
+Use .DATA <NNN> where <NNN> indicates the length of data in bytes if the
+response data is length-bounded. For instance, .DATA 1234 indicates that
+the following data is 1234 bytes long.
+
+Use .DATA chunked if you don't know the length of response data in advance.
+
+The actual data begins at the next line to .DATA.
+
+The length-bounded response message looks like this. The response message 
+handler must consume exactly the number of bytes specifed on the .LENGTH line
+starting from the beginning of the next line to .DATA without ignoring any 
+characters.
+
+```
+ .OK
+ .DATA 10
+ aaaaaaaaaa
+```
+
+The chunked data looks like this. Each chunk begins with the length and a colon.
+The 0 sized chunk indicates the end of data. A chunk size is in decimal and is
+followed by a colon. The actual data chunk starts immediately after the colon.
+The response processor must consume exactly the number of bytes specified in
+the chunk size part and attempt to read the size of the next chunk. 
+
+The whitespaces between the end of the previous chunk data and the next chunk
+size, if any, should get ignored.
+
+```
+ .OK
+ .DATA chunked
+ 4:xxxx10:abcdef
+ ghi
+ 0:
+```
+
+With the chunked data transfer format, you can revoke the ongoing response data
+and start a new response.
+
+```
+ .OK
+ .DATA chunked
+ 4:xxxx-1:.ERROR "error has occurred"
+```
+
+An optional attribute line is composed of the attribute name and the attribute
+value delimited by a whitespace. There are no defined attributes but the attribute
+name must not be one of .OK, ERROR, .DATA. The attribute value follows the same
+format as the status data of the short-form response.
+
+```
+ .OK
+ .TYPE json/utf8
+ .DATA chunked
+ ....
+```