Fix code review issues in MCP server

- Set analyzed flag in analyze() function instead of tool handlers
- Remove duplicate code in architecture documentation

Co-authored-by: michael-schwarz <13812333+michael-schwarz@users.noreply.github.com>

Author: Copilot

docs/developer-guide/mcp-architecture.md

@@ -76,10 +76,6 @@ Each tool follows this pattern:
   (* 4. Return result *)
   ToolResult.make_text "Success message"
 ```
-  
-  (* 4. Return result *)
-  ToolResult.make_text "Success message"
-```
 
 ## Configuration Handling
 

src/util/mcpServer.ml

@@ -187,6 +187,9 @@ let analyze ?(reset=false) (mcp_serv: t) =
   (* Run analysis *)
   Maingoblint.do_analyze None file;
   Maingoblint.do_gobview file;
+  
+  (* Mark as analyzed *)
+  mcp_serv.analyzed <- true
 
 (** JSON Schema helpers *)
 let string_schema = `Assoc [("type", `String "string")]
@@ -405,7 +408,6 @@ let handle_tools_call (mcp_serv: t) (call: ToolCall.t) =
 
         (* Run analysis directly *)
         analyze mcp_serv ~reset;
-        mcp_serv.analyzed <- true;
 
         let status = if !AnalysisState.verified = Some false then "VerifyError" else "Success" in
         let msg_count = List.length (Messages.Table.to_list ()) in
@@ -423,7 +425,6 @@ let handle_tools_call (mcp_serv: t) (call: ToolCall.t) =
 
         (* Run analysis directly *)
         analyze mcp_serv ~reset;
-        mcp_serv.analyzed <- true;
 
         let status = if !AnalysisState.verified = Some false then "VerifyError" else "Success" in
         let msg_count = List.length (Messages.Table.to_list ()) in