feat: read_messagesに本文範囲取得viewを追加HEADv0.1.2main

大きなメッセージをMCPホストのtool result上限に触れず取得できるよう、read_messagesへcontent viewを追加する。UTF-8 byte単位のcontent_offset/content_limitとnext_content_offsetにより、文字境界を壊さず反復取得できる。

各chunkにrevisionとtotal_bytesを返し、expected_revisionで取得途中の更新をCONFLICTとして検出する。未知または削除済みmessageはNOT_FOUND、不正な範囲はVALIDATION_ERRORとする。読み取りには既存memory indexを利用し、永続data、revision、I/O counterを変更しない。

新規MCP toolは増やさず、従来の7 tool構成とhistory/context/sessionsの契約を維持する。storage recordとsnapshot形式も変更しないためmigrationとrebuildは不要。

README、設計書、MCPリファレンスを更新し、多言語UTF-8本文の完全再構成、境界・limit・revision・削除済みmessage、read-only I/O、既存view回帰をprotocol testで検証する。

Tests: ctest --test-dir build/range-read-test -C Release --output-on-failure (20/20 passed)

1 files changed, 150 insertions, 2 deletions

diff --git a/src/mcp/tools/read_messages.c b/src/mcp/tools/read_messages.c
index 131c513..e217aa2 100644
--- a/src/mcp/tools/read_messages.c
+++ b/src/mcp/tools/read_messages.c
@@ -15,6 +15,8 @@
 #define READ_DEFAULT_LIMIT 100
 #define READ_MAX_LIMIT 1000
 #define READ_RESPONSE_BUDGET (8U * 1024U * 1024U - 64U * 1024U)
+#define CONTENT_RANGE_DEFAULT_LIMIT 4096
+#define CONTENT_RANGE_MAX_LIMIT 65536
 #define HISTORY_CURSOR_BYTES 47U
 #define HISTORY_CURSOR_TEXT_MAX 64U
 #define SESSION_CURSOR_BYTES_MAX (1U + 8U + 4U + TINY_SESSION_ID_MAX + 4U)
@@ -26,6 +28,120 @@ typedef struct history_cursor
   char message_id[27];
 } history_cursor_t;
 
+static bool
+utf8_boundary(const char *content, size_t length, size_t offset)
+{
+  return offset == 0 || offset == length || ((unsigned char)content[offset] & 0xc0U) != 0x80U;
+}
+
+static void
+call_content_view(tiny_mcp_tools_context_t *context, const mcpc_tool_t *tool, mcpc_ucbr_t *result)
+{
+  char *message_id = tiny_mcp_tool_get_string(tool, "message_id", TINY_MESSAGE_ID_LENGTH);
+  int32_t requested_offset = 0, requested_limit = CONTENT_RANGE_DEFAULT_LIMIT;
+  int32_t expected_revision = 0;
+  mcpc_errcode_t offset_status
+      = mcpc_tool_get_tpropval_i32(tool, (const char8_t *)"content_offset", &requested_offset);
+  mcpc_errcode_t limit_status
+      = mcpc_tool_get_tpropval_i32(tool, (const char8_t *)"content_limit", &requested_limit);
+  mcpc_errcode_t revision_status
+      = mcpc_tool_get_tpropval_i32(tool, (const char8_t *)"expected_revision", &expected_revision);
+  const tiny_store_entry_t *entry = NULL;
+  size_t entry_index = 0, content_length = 0, offset = 0, end = 0, chunk_length = 0;
+  char *chunk = NULL;
+  tiny_json_buffer_t json;
+  bool valid = context && message_id && tiny_message_id_valid(message_id)
+               && (offset_status == MCPC_EC_0 || offset_status == MCPC_EC_ANYDATA_NOTFOUND)
+               && requested_offset >= 0
+               && (limit_status == MCPC_EC_0 || limit_status == MCPC_EC_ANYDATA_NOTFOUND)
+               && requested_limit >= 1 && requested_limit <= CONTENT_RANGE_MAX_LIMIT
+               && (revision_status == MCPC_EC_0 || revision_status == MCPC_EC_ANYDATA_NOTFOUND)
+               && (revision_status != MCPC_EC_0 || expected_revision >= 1);
+  if (!valid)
+    {
+      tiny_mcp_tool_add_error(result,
+                              "VALIDATION_ERROR",
+                              "message_id, content_offset, content_limit or expected_revision is "
+                              "invalid");
+    }
+  else if (!tiny_mcp_tools_ensure_memory_index(context))
+    {
+      tiny_mcp_tool_add_error(result, "INTERNAL_ERROR", "message index allocation failed");
+    }
+  else if (!tiny_memory_index_find(context->memory_index, message_id, &entry_index)
+           || !(entry = tiny_store_entry(context->store, entry_index)) || entry->deleted)
+    {
+      tiny_mcp_tool_add_error(result, "NOT_FOUND", "message was not found");
+    }
+  else if (revision_status == MCPC_EC_0 && entry->message.revision != (uint64_t)expected_revision)
+    {
+      tiny_mcp_tool_add_error(result, "CONFLICT", "message revision does not match");
+    }
+  else
+    {
+      content_length = strlen(entry->message.content);
+      offset = (size_t)requested_offset;
+      if (offset > content_length || !utf8_boundary(entry->message.content, content_length, offset))
+        {
+          tiny_mcp_tool_add_error(result,
+                                  "VALIDATION_ERROR",
+                                  "content_offset is outside content or not a UTF-8 boundary");
+        }
+      else
+        {
+          end = offset + (size_t)requested_limit;
+          if (end > content_length)
+            {
+              end = content_length;
+            }
+          while (end > offset && !utf8_boundary(entry->message.content, content_length, end))
+            {
+              --end;
+            }
+          if (end == offset && end < content_length)
+            {
+              do
+                {
+                  ++end;
+                }
+              while (end < content_length
+                     && !utf8_boundary(entry->message.content, content_length, end));
+            }
+          chunk_length = end - offset;
+          chunk = malloc(chunk_length + 1);
+          if (!chunk)
+            {
+              tiny_mcp_tool_add_error(result, "INTERNAL_ERROR", "out of memory");
+            }
+          else
+            {
+              memcpy(chunk, entry->message.content + offset, chunk_length);
+              chunk[chunk_length] = '\0';
+              tiny_json_init(&json);
+              tiny_json_raw(&json, "{\"content_range\":{\"message_id\":");
+              tiny_json_string(&json, entry->message.message_id);
+              tiny_json_raw(&json, ",\"revision\":");
+              tiny_json_u64(&json, entry->message.revision);
+              tiny_json_raw(&json, ",\"total_bytes\":");
+              tiny_json_u64(&json, content_length);
+              tiny_json_raw(&json, ",\"content_offset\":");
+              tiny_json_u64(&json, offset);
+              tiny_json_raw(&json, ",\"next_content_offset\":");
+              tiny_json_u64(&json, end);
+              tiny_json_raw(&json, ",\"complete\":");
+              tiny_json_raw(&json, end == content_length ? "true" : "false");
+              tiny_json_raw(&json, ",\"content\":");
+              tiny_json_string(&json, chunk);
+              tiny_json_raw(&json, "}}");
+              tiny_mcp_tool_add_json(result, &json);
+              tiny_json_free(&json);
+            }
+        }
+    }
+  free(chunk);
+  free(message_id);
+}
+
 static uint64_t
 session_hash(const char *value)
 {
@@ -800,6 +916,14 @@ call_read_messages(const mcpc_tool_t *tool, mcpc_ucbr_t *result)
   const tiny_message_t *last = NULL;
   tiny_json_buffer_t json;
 
+  if (view && strcmp(view, "content") == 0)
+    {
+      call_content_view(context, tool, result);
+      free(view);
+      free(session);
+      free(cursor_text);
+      return;
+    }
   if (view && strcmp(view, "context") == 0)
     {
       call_context_view(context, tool, result);
@@ -950,14 +1074,38 @@ tiny_mcp_register_read_messages(mcpc_server_t *server)
                           call_read_messages);
   return tool
          && tiny_mcp_tool_add_property(
-             tool, "view", "Read view: history (default), context, or sessions.", MCPC_U8STR, false)
+             tool,
+             "view",
+             "Read view: history (default), content, context, or sessions.",
+             MCPC_U8STR,
+             false)
          && tiny_mcp_tool_add_property(tool,
                                        "session_id",
                                        "Conversation session whose chronological history to read.",
                                        MCPC_U8STR,
                                        false)
+         && tiny_mcp_tool_add_property(tool,
+                                       "message_id",
+                                       "Message ULID required by the content and context views.",
+                                       MCPC_U8STR,
+                                       false)
+         && tiny_mcp_tool_add_property(tool,
+                                       "content_offset",
+                                       "UTF-8 byte offset for the content view; defaults to 0.",
+                                       MCPC_I32,
+                                       false)
          && tiny_mcp_tool_add_property(
-             tool, "message_id", "Message ULID required by the context view.", MCPC_U8STR, false)
+             tool,
+             "content_limit",
+             "Target content chunk bytes from 1 to 65536; defaults to 4096.",
+             MCPC_I32,
+             false)
+         && tiny_mcp_tool_add_property(
+             tool,
+             "expected_revision",
+             "Optional positive revision for a consistent multi-call content read.",
+             MCPC_I32,
+             false)
          && tiny_mcp_tool_add_property(
              tool, "before", "Context messages before the target, from 0 to 10.", MCPC_I32, false)
          && tiny_mcp_tool_add_property(