LoginSignup
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

@t-akai

MCP Python SDKとGoogle ADKでMCPサーバ（ODBC経由PostgreSQL接続）を試してみた

Last updated at Posted at 2025-05-06

こんにちは。あかいです。
この記事は、勉強のためにMCPサーバを試した際の備忘録です。
2025/05/06時点で、Transport方法にStreamable HTTPが追加されていますが、MCP Python SDKで未実装のため、自前実装せずに旧SSEで試しています。
また、MCPサーバのresources、prompts、toolsのうち、toolsのみの検証です（adk webコマンドで起動するGUIからのresources等の参照方法を見つけられなかったため）。
その他の方針はこちら。

目次

MCPとは

2025-03-26(latest) @2025/05/06

概要（MCPプロトコル要求）

参考：Introducing the Model Context Protocol \ Anthropic

  • LLMにコンテキストを与えるためのオープンプロトコル（≠実装）

  • MCPホスト、MCPクライアント、MCPサーバからなる
    image.png

    • MCP Hosts: Programs like Claude Desktop, IDEs, or AI tools that want to access data through MCP
    • MCP Clients: Protocol clients that maintain 1:1 connections with servers
    • MCP Servers: Lightweight programs that each expose specific capabilities through the standardized Model Context Protocol
    • Local Data Sources: Your computer’s files, databases, and services that MCP servers can securely access
    • Remote Services: External systems available over the internet (e.g., through APIs) that MCP servers can connect to

  • ベースとなるプロトコル: JSON-RPC 2.0

    ※ バッチ処理可
    ※ UTF-8

    • Request(クライアント<->サーバ)

      {
    jsonrpc: "2.0";
    id: string | number;
    method: string;
    params?: {
        [key: string]: unknown;
    };
}

    • Response(クライアント<->サーバ)

      {
    jsonrpc: "2.0";
    id: string | number;
    result?: {
        [key: string]: unknown;
    }
    error?: {
        code: number;
        message: string;
        data?: unknown;
    }
}

    • Notification(クライアント->サーバ、サーバ->クライアント)

      {
    jsonrpc: "2.0";
    method: string;
    params?: {
        [key: string]: unknown;
    };
}

  • トランスポート
    参考: https://zenn.dev/atsukish/articles/e4e1009ab19c3c

    • 2種の通信方式
      • STDIO
        • MCPクライアントがMCPサーバをサブプロセスとして起動
        • STDIN/STDOUTを使用
        • メッセージは改行区切り
      • Streamable HTTP
        • サーバは1つのHTTPエンドポイントを用意
        • POST、GETを使用
          • POST: JSON-RPCメッセージの送信（クライアント<->サーバ）
          • GET: SSEストリームの開始要求（クライアント->サーバ）
        • レスポンスコード
          • 202: 受付
          • 405: SSEストリーム非対応
          • その他エラーコード: その他エラー時
        • SSE (Server-Sent Events)ストリーム
          • クライアント、サーバ両方からいつでも閉じてよい
          • クライアント:サーバ = 多:多
          • ネットワーク切断後に再開可能
      • (SSE):
        • 古いプロトコル
        • SSE用エンドポイントと、POST用エンドポイントを設ける

  • 認証・認可

    • HTTP利用時: HTTPベースの認証
    • STDIO利用時: 環境からクレデンシャルを取得

  • スキーマ

    Typescriptスキーマとして定義される
    (TypescriptからJSONスキーマが自動生成されるが)

  • ライフサイクル

    1. Initialization: Capability negotiation and protocol version agreement
    2. Operation: Normal protocol communication
    3. Shutdown: Graceful termination of the connection

    image.png

  • エラーハンドリング

    • Error response example

      {
    "jsonrpc": "2.0",
    "id": 1,
    "error": {
        "code": -32602,
        "message": "Unsupported protocol version",
        "data": {
        "supported": ["2024-11-05"],
        "requested": "1.0.0"
        }
    }
}

実装

クライアント実装

クライアント名 実装言語 対応接続方式 ホスト形態 備考
LangChain Python STDIO, SSE ローカル langchain-mcp-tools を使用
LangChain.js TypeScript STDIO, SSE ローカル langchain-mcp-adapters により複数接続可
Claude Desktop Electron STDIO ローカル GUIアプリ、ローカルMCPとの接続設定あり
Sourcegraph Cody Go STDIO, SSE ローカル エディタ統合型
Zed Editor Rust STDIO ローカル Rust製エディタ、内部でMCPを使用
Cursor TypeScript STDIO ローカル VSCode風エディタ、プラグイン経由で接続
Continue TypeScript STDIO, SSE ローカル 複数MCPサーバー対応
LibreChat JavaScript STDIO, SSE ローカル／自己ホスト MCPツール統合可能。Dockerで自己ホストも簡易
OpenAI Agents SDK Python STDIO, SSE ローカル MCP対応のOpenAIクライアント
PydanticAI Python STDIO, SSE ローカル Python統合用
Google ADK Python STDIO, SSE ローカル／自己ホスト MCPToolsetで接続。Gemma等と組み合わせてエージェント開発可能

※ ChatGPT 4oで生成。詳細未確認。
※ 自己ホストとは、他者にも使える前提でLAN/インターネットで公開可能なものとのこと（ポートを開けるとか、localhost限定しないとか）。

サーバ実装

名称 言語 接続方式 ホスト形態 提供元・備考
MCP Python SDK Python STDIO, SSE ローカル／自己ホスト Anthropic公式。FastAPI等と組み合わせて使用可能。
MCP TypeScript SDK TypeScript STDIO, SSE ローカル／自己ホスト Node.js環境向け。
MCP Java SDK Java STDIO, SSE ローカル／自己ホスト Spring Bootなどの環境で利用可能。
MCP Kotlin SDK Kotlin STDIO, SSE ローカル／自己ホスト JetBrains協力。クライアント・サーバ両対応。
MCP C# SDK C# STDIO, SSE ローカル／自己ホスト Microsoft協力。NuGet経由で提供。
MCP Swift SDK Swift STDIO, SSE ローカル／自己ホスト Appleプラットフォーム向け。
MCP Rust SDK Rust STDIO, SSE ローカル／自己ホスト コミュニティ主導。
EasyMCP TypeScript STDIO ローカル／自己ホスト Express風の構成でMCPツールを簡易実装可能。
FastAPI-MCP Python SSE ローカル／自己ホスト FastAPIを活用しAPI公開型のMCPを実装。
LangServe Python SSE ローカル／自己ホスト LangChain製。LangChainツールをMCP化可能。
Azure MCP Server 任意 SSE SaaS Microsoft提供。AzureベースのMCP公開構成。
Composio MCP 不明 SSE SaaS SaaS型MCP集約サービス。OAuth統合が豊富。

※ ChatGPT 4oで生成。詳細未確認。
※ MCP Python SDKには、FastMCPが取り込まれている。別途、FastMCP2.0も公開されている。

お試し

方針

image.png

今回は、Ubuntu上に、PostgreSQLのMCPサーバをお試しで構築する。
MCPサーバ、クライアント間はHTTP接続とする。ただし、MCP Python SDKでのStreamable HTTPの実装が手間なので、 旧SSEで実装する ※。

※ 実装例が公開されているが、まだMCP Python SDKでは対応していないよう。

使用するライブラリは以下の通り。

構築

依存関係

Ubuntu          22.04
Python          3.12.10
  uv            0.7.2
    google-adk  0.4.0
    litellm     1.68.0
    mcp[cli]    1.7.1
    pyodbc      5.2.0
postgresql      14.17-0ubuntu0.22.04.1
odbc-postgresql 1:13.02.0000-2

ディレクトリ構成

構築手順後の状態

mcp_test
├── README.md
├── agent_samples
│   ├── .env
│   └── agent
│       ├── __init__.py
│       └── agent.py
├── server_samples
│   └── server.py
├── pyproject.toml
└── uv.lock

構築手順

  1. データ準備

    e-Statからデータを取得。
    今回は、男女別人口（各年10月1日現在）- 総人口、日本人人口（2000年～2020年）を使用。
    マトリクス表なので、Excelで開いてPowerQueryエディタからアンピボットしておく。

    データ
    population_2000-2020.csv
    ID,prefecture,year,population
1,Hokkaido,2000,5683
2,Hokkaido,2001,5680
3,Hokkaido,2002,5672
4,Hokkaido,2003,5663
5,Hokkaido,2004,5650
6,Hokkaido,2005,5628
7,Hokkaido,2006,5605
8,Hokkaido,2007,5579
9,Hokkaido,2008,5548
10,Hokkaido,2009,5524
11,Hokkaido,2010,5506
12,Hokkaido,2011,5488
13,Hokkaido,2012,5465
14,Hokkaido,2013,5438
15,Hokkaido,2014,5410
16,Hokkaido,2015,5382
17,Hokkaido,2016,5355
18,Hokkaido,2017,5325
19,Hokkaido,2018,5293
20,Hokkaido,2019,5259
21,Hokkaido,2020,5225
22,Aomori-ken,2000,1476
23,Aomori-ken,2001,1473
24,Aomori-ken,2002,1467
25,Aomori-ken,2003,1459
26,Aomori-ken,2004,1448
27,Aomori-ken,2005,1437
28,Aomori-ken,2006,1424
29,Aomori-ken,2007,1409
30,Aomori-ken,2008,1395
31,Aomori-ken,2009,1383
32,Aomori-ken,2010,1373
33,Aomori-ken,2011,1363
34,Aomori-ken,2012,1350
35,Aomori-ken,2013,1337
36,Aomori-ken,2014,1323
37,Aomori-ken,2015,1308
38,Aomori-ken,2016,1295
39,Aomori-ken,2017,1282
40,Aomori-ken,2018,1268
41,Aomori-ken,2019,1253
42,Aomori-ken,2020,1238
43,Iwate-ken,2000,1416
44,Iwate-ken,2001,1413
45,Iwate-ken,2002,1407
46,Iwate-ken,2003,1401
47,Iwate-ken,2004,1395
48,Iwate-ken,2005,1385
49,Iwate-ken,2006,1375
50,Iwate-ken,2007,1364
51,Iwate-ken,2008,1352
52,Iwate-ken,2009,1340
53,Iwate-ken,2010,1330
54,Iwate-ken,2011,1315
55,Iwate-ken,2012,1306
56,Iwate-ken,2013,1299
57,Iwate-ken,2014,1290
58,Iwate-ken,2015,1280
59,Iwate-ken,2016,1268
60,Iwate-ken,2017,1254
61,Iwate-ken,2018,1240
62,Iwate-ken,2019,1226
63,Iwate-ken,2020,1211
64,Miyagi-ken,2000,2365
65,Miyagi-ken,2001,2369
66,Miyagi-ken,2002,2369
67,Miyagi-ken,2003,2369
68,Miyagi-ken,2004,2366
69,Miyagi-ken,2005,2360
70,Miyagi-ken,2006,2358
71,Miyagi-ken,2007,2354
72,Miyagi-ken,2008,2349
73,Miyagi-ken,2009,2348
74,Miyagi-ken,2010,2348
75,Miyagi-ken,2011,2326
76,Miyagi-ken,2012,2329
77,Miyagi-ken,2013,2333
78,Miyagi-ken,2014,2335
79,Miyagi-ken,2015,2334
80,Miyagi-ken,2016,2332
81,Miyagi-ken,2017,2326
82,Miyagi-ken,2018,2320
83,Miyagi-ken,2019,2312
84,Miyagi-ken,2020,2302
85,Akita-ken,2000,1189
86,Akita-ken,2001,1183
87,Akita-ken,2002,1175
88,Akita-ken,2003,1165
89,Akita-ken,2004,1156
90,Akita-ken,2005,1146
91,Akita-ken,2006,1134
92,Akita-ken,2007,1121
93,Akita-ken,2008,1109
94,Akita-ken,2009,1097
95,Akita-ken,2010,1086
96,Akita-ken,2011,1075
97,Akita-ken,2012,1063
98,Akita-ken,2013,1050
99,Akita-ken,2014,1037
100,Akita-ken,2015,1023
101,Akita-ken,2016,1011
102,Akita-ken,2017,999
103,Akita-ken,2018,985
104,Akita-ken,2019,972
105,Akita-ken,2020,960
106,Yamagata-ken,2000,1244
107,Yamagata-ken,2001,1241
108,Yamagata-ken,2002,1236
109,Yamagata-ken,2003,1230
110,Yamagata-ken,2004,1224
111,Yamagata-ken,2005,1216
112,Yamagata-ken,2006,1207
113,Yamagata-ken,2007,1198
114,Yamagata-ken,2008,1188
115,Yamagata-ken,2009,1178
116,Yamagata-ken,2010,1169
117,Yamagata-ken,2011,1162
118,Yamagata-ken,2012,1153
119,Yamagata-ken,2013,1144
120,Yamagata-ken,2014,1134
121,Yamagata-ken,2015,1124
122,Yamagata-ken,2016,1114
123,Yamagata-ken,2017,1103
124,Yamagata-ken,2018,1092
125,Yamagata-ken,2019,1080
126,Yamagata-ken,2020,1068
127,Fukushima-ken,2000,2127
128,Fukushima-ken,2001,2124
129,Fukushima-ken,2002,2118
130,Fukushima-ken,2003,2110
131,Fukushima-ken,2004,2102
132,Fukushima-ken,2005,2091
133,Fukushima-ken,2006,2080
134,Fukushima-ken,2007,2067
135,Fukushima-ken,2008,2054
136,Fukushima-ken,2009,2041
137,Fukushima-ken,2010,2029
138,Fukushima-ken,2011,1988
139,Fukushima-ken,2012,1957
140,Fukushima-ken,2013,1940
141,Fukushima-ken,2014,1927
142,Fukushima-ken,2015,1914
143,Fukushima-ken,2016,1903
144,Fukushima-ken,2017,1886
145,Fukushima-ken,2018,1869
146,Fukushima-ken,2019,1852
147,Fukushima-ken,2020,1833
148,Ibaraki-ken,2000,2986
149,Ibaraki-ken,2001,2990
150,Ibaraki-ken,2002,2987
151,Ibaraki-ken,2003,2985
152,Ibaraki-ken,2004,2982
153,Ibaraki-ken,2005,2975
154,Ibaraki-ken,2006,2974
155,Ibaraki-ken,2007,2973
156,Ibaraki-ken,2008,2971
157,Ibaraki-ken,2009,2970
158,Ibaraki-ken,2010,2970
159,Ibaraki-ken,2011,2960
160,Ibaraki-ken,2012,2947
161,Ibaraki-ken,2013,2937
162,Ibaraki-ken,2014,2927
163,Ibaraki-ken,2015,2917
164,Ibaraki-ken,2016,2910
165,Ibaraki-ken,2017,2902
166,Ibaraki-ken,2018,2892
167,Ibaraki-ken,2019,2879
168,Ibaraki-ken,2020,2867
169,Tochigi-ken,2000,2005
170,Tochigi-ken,2001,2011
171,Tochigi-ken,2002,2012
172,Tochigi-ken,2003,2014
173,Tochigi-ken,2004,2016
174,Tochigi-ken,2005,2017
175,Tochigi-ken,2006,2016
176,Tochigi-ken,2007,2016
177,Tochigi-ken,2008,2015
178,Tochigi-ken,2009,2011
179,Tochigi-ken,2010,2008
180,Tochigi-ken,2011,2000
181,Tochigi-ken,2012,1992
182,Tochigi-ken,2013,1986
183,Tochigi-ken,2014,1980
184,Tochigi-ken,2015,1974
185,Tochigi-ken,2016,1969
186,Tochigi-ken,2017,1962
187,Tochigi-ken,2018,1953
188,Tochigi-ken,2019,1943
189,Tochigi-ken,2020,1933
190,Gumma-ken,2000,2025
191,Gumma-ken,2001,2029
192,Gumma-ken,2002,2030
193,Gumma-ken,2003,2030
194,Gumma-ken,2004,2027
195,Gumma-ken,2005,2024
196,Gumma-ken,2006,2022
197,Gumma-ken,2007,2020
198,Gumma-ken,2008,2017
199,Gumma-ken,2009,2014
200,Gumma-ken,2010,2008
201,Gumma-ken,2011,2001
202,Gumma-ken,2012,1994
203,Gumma-ken,2013,1986
204,Gumma-ken,2014,1979
205,Gumma-ken,2015,1973
206,Gumma-ken,2016,1969
207,Gumma-ken,2017,1963
208,Gumma-ken,2018,1957
209,Gumma-ken,2019,1949
210,Gumma-ken,2020,1939
211,Saitama-ken,2000,6938
212,Saitama-ken,2001,6977
213,Saitama-ken,2002,7000
214,Saitama-ken,2003,7028
215,Saitama-ken,2004,7046
216,Saitama-ken,2005,7054
217,Saitama-ken,2006,7079
218,Saitama-ken,2007,7106
219,Saitama-ken,2008,7136
220,Saitama-ken,2009,7161
221,Saitama-ken,2010,7195
222,Saitama-ken,2011,7209
223,Saitama-ken,2012,7216
224,Saitama-ken,2013,7228
225,Saitama-ken,2014,7247
226,Saitama-ken,2015,7267
227,Saitama-ken,2016,7288
228,Saitama-ken,2017,7307
229,Saitama-ken,2018,7325
230,Saitama-ken,2019,7342
231,Saitama-ken,2020,7345
232,Chiba-ken,2000,5926
233,Chiba-ken,2001,5970
234,Chiba-ken,2002,5998
235,Chiba-ken,2003,6030
236,Chiba-ken,2004,6047
237,Chiba-ken,2005,6056
238,Chiba-ken,2006,6084
239,Chiba-ken,2007,6119
240,Chiba-ken,2008,6153
241,Chiba-ken,2009,6180
242,Chiba-ken,2010,6216
243,Chiba-ken,2011,6217
244,Chiba-ken,2012,6200
245,Chiba-ken,2013,6201
246,Chiba-ken,2014,6209
247,Chiba-ken,2015,6223
248,Chiba-ken,2016,6242
249,Chiba-ken,2017,6258
250,Chiba-ken,2018,6273
251,Chiba-ken,2019,6283
252,Chiba-ken,2020,6284
253,Tokyo-to,2000,12064
254,Tokyo-to,2001,12165
255,Tokyo-to,2002,12271
256,Tokyo-to,2003,12388
257,Tokyo-to,2004,12482
258,Tokyo-to,2005,12577
259,Tokyo-to,2006,12704
260,Tokyo-to,2007,12848
261,Tokyo-to,2008,12973
262,Tokyo-to,2009,13048
263,Tokyo-to,2010,13159
264,Tokyo-to,2011,13198
265,Tokyo-to,2012,13234
266,Tokyo-to,2013,13307
267,Tokyo-to,2014,13399
268,Tokyo-to,2015,13515
269,Tokyo-to,2016,13646
270,Tokyo-to,2017,13768
271,Tokyo-to,2018,13887
272,Tokyo-to,2019,14007
273,Tokyo-to,2020,14048
274,Kanagawa-ken,2000,8490
275,Kanagawa-ken,2001,8575
276,Kanagawa-ken,2002,8636
277,Kanagawa-ken,2003,8702
278,Kanagawa-ken,2004,8753
279,Kanagawa-ken,2005,8792
280,Kanagawa-ken,2006,8846
281,Kanagawa-ken,2007,8912
282,Kanagawa-ken,2008,8965
283,Kanagawa-ken,2009,9006
284,Kanagawa-ken,2010,9048
285,Kanagawa-ken,2011,9060
286,Kanagawa-ken,2012,9070
287,Kanagawa-ken,2013,9084
288,Kanagawa-ken,2014,9103
289,Kanagawa-ken,2015,9126
290,Kanagawa-ken,2016,9152
291,Kanagawa-ken,2017,9173
292,Kanagawa-ken,2018,9197
293,Kanagawa-ken,2019,9224
294,Kanagawa-ken,2020,9237
295,Niigata-ken,2000,2476
296,Niigata-ken,2001,2471
297,Niigata-ken,2002,2462
298,Niigata-ken,2003,2455
299,Niigata-ken,2004,2445
300,Niigata-ken,2005,2431
301,Niigata-ken,2006,2420
302,Niigata-ken,2007,2408
303,Niigata-ken,2008,2396
304,Niigata-ken,2009,2385
305,Niigata-ken,2010,2374
306,Niigata-ken,2011,2364
307,Niigata-ken,2012,2350
308,Niigata-ken,2013,2336
309,Niigata-ken,2014,2320
310,Niigata-ken,2015,2304
311,Niigata-ken,2016,2286
312,Niigata-ken,2017,2267
313,Niigata-ken,2018,2246
314,Niigata-ken,2019,2224
315,Niigata-ken,2020,2201
316,Toyama-ken,2000,1121
317,Toyama-ken,2001,1120
318,Toyama-ken,2002,1118
319,Toyama-ken,2003,1116
320,Toyama-ken,2004,1115
321,Toyama-ken,2005,1112
322,Toyama-ken,2006,1110
323,Toyama-ken,2007,1107
324,Toyama-ken,2008,1103
325,Toyama-ken,2009,1098
326,Toyama-ken,2010,1093
327,Toyama-ken,2011,1088
328,Toyama-ken,2012,1083
329,Toyama-ken,2013,1077
330,Toyama-ken,2014,1072
331,Toyama-ken,2015,1066
332,Toyama-ken,2016,1061
333,Toyama-ken,2017,1056
334,Toyama-ken,2018,1050
335,Toyama-ken,2019,1043
336,Toyama-ken,2020,1035
337,Ishikawa-ken,2000,1181
338,Ishikawa-ken,2001,1182
339,Ishikawa-ken,2002,1180
340,Ishikawa-ken,2003,1179
341,Ishikawa-ken,2004,1178
342,Ishikawa-ken,2005,1174
343,Ishikawa-ken,2006,1173
344,Ishikawa-ken,2007,1173
345,Ishikawa-ken,2008,1172
346,Ishikawa-ken,2009,1171
347,Ishikawa-ken,2010,1170
348,Ishikawa-ken,2011,1167
349,Ishikawa-ken,2012,1164
350,Ishikawa-ken,2013,1160
351,Ishikawa-ken,2014,1157
352,Ishikawa-ken,2015,1154
353,Ishikawa-ken,2016,1151
354,Ishikawa-ken,2017,1148
355,Ishikawa-ken,2018,1145
356,Ishikawa-ken,2019,1139
357,Ishikawa-ken,2020,1133
358,Fukui-ken,2000,829
359,Fukui-ken,2001,830
360,Fukui-ken,2002,828
361,Fukui-ken,2003,827
362,Fukui-ken,2004,824
363,Fukui-ken,2005,822
364,Fukui-ken,2006,820
365,Fukui-ken,2007,817
366,Fukui-ken,2008,814
367,Fukui-ken,2009,810
368,Fukui-ken,2010,806
369,Fukui-ken,2011,803
370,Fukui-ken,2012,800
371,Fukui-ken,2013,796
372,Fukui-ken,2014,791
373,Fukui-ken,2015,787
374,Fukui-ken,2016,783
375,Fukui-ken,2017,780
376,Fukui-ken,2018,777
377,Fukui-ken,2019,771
378,Fukui-ken,2020,767
379,Yamanashi-ken,2000,888
380,Yamanashi-ken,2001,890
381,Yamanashi-ken,2002,889
382,Yamanashi-ken,2003,888
383,Yamanashi-ken,2004,886
384,Yamanashi-ken,2005,885
385,Yamanashi-ken,2006,880
386,Yamanashi-ken,2007,877
387,Yamanashi-ken,2008,871
388,Yamanashi-ken,2009,867
389,Yamanashi-ken,2010,863
390,Yamanashi-ken,2011,857
391,Yamanashi-ken,2012,852
392,Yamanashi-ken,2013,847
393,Yamanashi-ken,2014,841
394,Yamanashi-ken,2015,835
395,Yamanashi-ken,2016,831
396,Yamanashi-ken,2017,826
397,Yamanashi-ken,2018,821
398,Yamanashi-ken,2019,815
399,Yamanashi-ken,2020,810
400,Nagano-ken,2000,2215
401,Nagano-ken,2001,2222
402,Nagano-ken,2002,2215
403,Nagano-ken,2003,2210
404,Nagano-ken,2004,2206
405,Nagano-ken,2005,2196
406,Nagano-ken,2006,2189
407,Nagano-ken,2007,2182
408,Nagano-ken,2008,2173
409,Nagano-ken,2009,2162
410,Nagano-ken,2010,2152
411,Nagano-ken,2011,2142
412,Nagano-ken,2012,2132
413,Nagano-ken,2013,2122
414,Nagano-ken,2014,2110
415,Nagano-ken,2015,2099
416,Nagano-ken,2016,2091
417,Nagano-ken,2017,2082
418,Nagano-ken,2018,2073
419,Nagano-ken,2019,2061
420,Nagano-ken,2020,2048
421,Gifu-ken,2000,2108
422,Gifu-ken,2001,2111
423,Gifu-ken,2002,2110
424,Gifu-ken,2003,2110
425,Gifu-ken,2004,2108
426,Gifu-ken,2005,2107
427,Gifu-ken,2006,2105
428,Gifu-ken,2007,2104
429,Gifu-ken,2008,2100
430,Gifu-ken,2009,2091
431,Gifu-ken,2010,2081
432,Gifu-ken,2011,2071
433,Gifu-ken,2012,2062
434,Gifu-ken,2013,2053
435,Gifu-ken,2014,2043
436,Gifu-ken,2015,2032
437,Gifu-ken,2016,2024
438,Gifu-ken,2017,2012
439,Gifu-ken,2018,2001
440,Gifu-ken,2019,1992
441,Gifu-ken,2020,1979
442,Shizuoka-ken,2000,3767
443,Shizuoka-ken,2001,3780
444,Shizuoka-ken,2002,3784
445,Shizuoka-ken,2003,3790
446,Shizuoka-ken,2004,3791
447,Shizuoka-ken,2005,3792
448,Shizuoka-ken,2006,3795
449,Shizuoka-ken,2007,3796
450,Shizuoka-ken,2008,3793
451,Shizuoka-ken,2009,3783
452,Shizuoka-ken,2010,3765
453,Shizuoka-ken,2011,3752
454,Shizuoka-ken,2012,3739
455,Shizuoka-ken,2013,3730
456,Shizuoka-ken,2014,3715
457,Shizuoka-ken,2015,3700
458,Shizuoka-ken,2016,3690
459,Shizuoka-ken,2017,3681
460,Shizuoka-ken,2018,3667
461,Shizuoka-ken,2019,3653
462,Shizuoka-ken,2020,3633
463,Aichi-ken,2000,7043
464,Aichi-ken,2001,7091
465,Aichi-ken,2002,7131
466,Aichi-ken,2003,7170
467,Aichi-ken,2004,7209
468,Aichi-ken,2005,7255
469,Aichi-ken,2006,7307
470,Aichi-ken,2007,7357
471,Aichi-ken,2008,7399
472,Aichi-ken,2009,7411
473,Aichi-ken,2010,7411
474,Aichi-ken,2011,7418
475,Aichi-ken,2012,7431
476,Aichi-ken,2013,7449
477,Aichi-ken,2014,7464
478,Aichi-ken,2015,7483
479,Aichi-ken,2016,7509
480,Aichi-ken,2017,7528
481,Aichi-ken,2018,7541
482,Aichi-ken,2019,7557
483,Aichi-ken,2020,7542
484,Mie-ken,2000,1857
485,Mie-ken,2001,1862
486,Mie-ken,2002,1863
487,Mie-ken,2003,1864
488,Mie-ken,2004,1867
489,Mie-ken,2005,1867
490,Mie-ken,2006,1872
491,Mie-ken,2007,1873
492,Mie-ken,2008,1871
493,Mie-ken,2009,1864
494,Mie-ken,2010,1855
495,Mie-ken,2011,1847
496,Mie-ken,2012,1841
497,Mie-ken,2013,1833
498,Mie-ken,2014,1826
499,Mie-ken,2015,1816
500,Mie-ken,2016,1809
501,Mie-ken,2017,1801
502,Mie-ken,2018,1793
503,Mie-ken,2019,1783
504,Mie-ken,2020,1770
505,Shiga-ken,2000,1343
506,Shiga-ken,2001,1354
507,Shiga-ken,2002,1361
508,Shiga-ken,2003,1368
509,Shiga-ken,2004,1375
510,Shiga-ken,2005,1380
511,Shiga-ken,2006,1390
512,Shiga-ken,2007,1398
513,Shiga-ken,2008,1405
514,Shiga-ken,2009,1409
515,Shiga-ken,2010,1411
516,Shiga-ken,2011,1413
517,Shiga-ken,2012,1414
518,Shiga-ken,2013,1415
519,Shiga-ken,2014,1414
520,Shiga-ken,2015,1413
521,Shiga-ken,2016,1414
522,Shiga-ken,2017,1414
523,Shiga-ken,2018,1414
524,Shiga-ken,2019,1416
525,Shiga-ken,2020,1414
526,Kyoto-fu,2000,2644
527,Kyoto-fu,2001,2649
528,Kyoto-fu,2002,2648
529,Kyoto-fu,2003,2650
530,Kyoto-fu,2004,2650
531,Kyoto-fu,2005,2648
532,Kyoto-fu,2006,2646
533,Kyoto-fu,2007,2643
534,Kyoto-fu,2008,2640
535,Kyoto-fu,2009,2637
536,Kyoto-fu,2010,2636
537,Kyoto-fu,2011,2633
538,Kyoto-fu,2012,2628
539,Kyoto-fu,2013,2622
540,Kyoto-fu,2014,2616
541,Kyoto-fu,2015,2610
542,Kyoto-fu,2016,2608
543,Kyoto-fu,2017,2604
544,Kyoto-fu,2018,2598
545,Kyoto-fu,2019,2592
546,Kyoto-fu,2020,2578
547,Osaka-fu,2000,8805
548,Osaka-fu,2001,8821
549,Osaka-fu,2002,8821
550,Osaka-fu,2003,8824
551,Osaka-fu,2004,8825
552,Osaka-fu,2005,8817
553,Osaka-fu,2006,8828
554,Osaka-fu,2007,8839
555,Osaka-fu,2008,8847
556,Osaka-fu,2009,8855
557,Osaka-fu,2010,8865
558,Osaka-fu,2011,8863
559,Osaka-fu,2012,8861
560,Osaka-fu,2013,8856
561,Osaka-fu,2014,8845
562,Osaka-fu,2015,8839
563,Osaka-fu,2016,8841
564,Osaka-fu,2017,8841
565,Osaka-fu,2018,8838
566,Osaka-fu,2019,8842
567,Osaka-fu,2020,8838
568,Hyogo-ken,2000,5551
569,Hyogo-ken,2001,5572
570,Hyogo-ken,2002,5580
571,Hyogo-ken,2003,5589
572,Hyogo-ken,2004,5592
573,Hyogo-ken,2005,5591
574,Hyogo-ken,2006,5592
575,Hyogo-ken,2007,5593
576,Hyogo-ken,2008,5592
577,Hyogo-ken,2009,5591
578,Hyogo-ken,2010,5588
579,Hyogo-ken,2011,5584
580,Hyogo-ken,2012,5575
581,Hyogo-ken,2013,5565
582,Hyogo-ken,2014,5550
583,Hyogo-ken,2015,5535
584,Hyogo-ken,2016,5526
585,Hyogo-ken,2017,5515
586,Hyogo-ken,2018,5501
587,Hyogo-ken,2019,5488
588,Hyogo-ken,2020,5465
589,Nara-ken,2000,1443
590,Nara-ken,2001,1442
591,Nara-ken,2002,1437
592,Nara-ken,2003,1434
593,Nara-ken,2004,1428
594,Nara-ken,2005,1421
595,Nara-ken,2006,1417
596,Nara-ken,2007,1413
597,Nara-ken,2008,1407
598,Nara-ken,2009,1404
599,Nara-ken,2010,1401
600,Nara-ken,2011,1395
601,Nara-ken,2012,1388
602,Nara-ken,2013,1381
603,Nara-ken,2014,1373
604,Nara-ken,2015,1364
605,Nara-ken,2016,1357
606,Nara-ken,2017,1349
607,Nara-ken,2018,1341
608,Nara-ken,2019,1333
609,Nara-ken,2020,1324
610,Wakayama-ken,2000,1070
611,Wakayama-ken,2001,1065
612,Wakayama-ken,2002,1059
613,Wakayama-ken,2003,1052
614,Wakayama-ken,2004,1045
615,Wakayama-ken,2005,1036
616,Wakayama-ken,2006,1029
617,Wakayama-ken,2007,1021
618,Wakayama-ken,2008,1014
619,Wakayama-ken,2009,1008
620,Wakayama-ken,2010,1002
621,Wakayama-ken,2011,995
622,Wakayama-ken,2012,988
623,Wakayama-ken,2013,980
624,Wakayama-ken,2014,972
625,Wakayama-ken,2015,964
626,Wakayama-ken,2016,956
627,Wakayama-ken,2017,948
628,Wakayama-ken,2018,940
629,Wakayama-ken,2019,931
630,Wakayama-ken,2020,923
631,Tottori-ken,2000,613
632,Tottori-ken,2001,613
633,Tottori-ken,2002,612
634,Tottori-ken,2003,611
635,Tottori-ken,2004,610
636,Tottori-ken,2005,607
637,Tottori-ken,2006,604
638,Tottori-ken,2007,601
639,Tottori-ken,2008,596
640,Tottori-ken,2009,592
641,Tottori-ken,2010,589
642,Tottori-ken,2011,586
643,Tottori-ken,2012,583
644,Tottori-ken,2013,580
645,Tottori-ken,2014,577
646,Tottori-ken,2015,573
647,Tottori-ken,2016,570
648,Tottori-ken,2017,566
649,Tottori-ken,2018,562
650,Tottori-ken,2019,557
651,Tottori-ken,2020,553
652,Shimane-ken,2000,762
653,Shimane-ken,2001,760
654,Shimane-ken,2002,756
655,Shimane-ken,2003,752
656,Shimane-ken,2004,747
657,Shimane-ken,2005,742
658,Shimane-ken,2006,737
659,Shimane-ken,2007,733
660,Shimane-ken,2008,727
661,Shimane-ken,2009,721
662,Shimane-ken,2010,717
663,Shimane-ken,2011,713
664,Shimane-ken,2012,708
665,Shimane-ken,2013,704
666,Shimane-ken,2014,699
667,Shimane-ken,2015,694
668,Shimane-ken,2016,691
669,Shimane-ken,2017,687
670,Shimane-ken,2018,682
671,Shimane-ken,2019,677
672,Shimane-ken,2020,671
673,Okayama-ken,2000,1951
674,Okayama-ken,2001,1954
675,Okayama-ken,2002,1956
676,Okayama-ken,2003,1958
677,Okayama-ken,2004,1959
678,Okayama-ken,2005,1957
679,Okayama-ken,2006,1956
680,Okayama-ken,2007,1956
681,Okayama-ken,2008,1953
682,Okayama-ken,2009,1949
683,Okayama-ken,2010,1945
684,Okayama-ken,2011,1941
685,Okayama-ken,2012,1937
686,Okayama-ken,2013,1932
687,Okayama-ken,2014,1926
688,Okayama-ken,2015,1922
689,Okayama-ken,2016,1917
690,Okayama-ken,2017,1911
691,Okayama-ken,2018,1904
692,Okayama-ken,2019,1897
693,Okayama-ken,2020,1888
694,Hiroshima-ken,2000,2879
695,Hiroshima-ken,2001,2880
696,Hiroshima-ken,2002,2878
697,Hiroshima-ken,2003,2879
698,Hiroshima-ken,2004,2879
699,Hiroshima-ken,2005,2877
700,Hiroshima-ken,2006,2875
701,Hiroshima-ken,2007,2874
702,Hiroshima-ken,2008,2870
703,Hiroshima-ken,2009,2864
704,Hiroshima-ken,2010,2861
705,Hiroshima-ken,2011,2858
706,Hiroshima-ken,2012,2855
707,Hiroshima-ken,2013,2850
708,Hiroshima-ken,2014,2846
709,Hiroshima-ken,2015,2844
710,Hiroshima-ken,2016,2840
711,Hiroshima-ken,2017,2833
712,Hiroshima-ken,2018,2824
713,Hiroshima-ken,2019,2813
714,Hiroshima-ken,2020,2800
715,Yamaguchi-ken,2000,1528
716,Yamaguchi-ken,2001,1523
717,Yamaguchi-ken,2002,1517
718,Yamaguchi-ken,2003,1510
719,Yamaguchi-ken,2004,1502
720,Yamaguchi-ken,2005,1493
721,Yamaguchi-ken,2006,1484
722,Yamaguchi-ken,2007,1476
723,Yamaguchi-ken,2008,1466
724,Yamaguchi-ken,2009,1459
725,Yamaguchi-ken,2010,1451
726,Yamaguchi-ken,2011,1444
727,Yamaguchi-ken,2012,1434
728,Yamaguchi-ken,2013,1425
729,Yamaguchi-ken,2014,1415
730,Yamaguchi-ken,2015,1405
731,Yamaguchi-ken,2016,1394
732,Yamaguchi-ken,2017,1382
733,Yamaguchi-ken,2018,1369
734,Yamaguchi-ken,2019,1357
735,Yamaguchi-ken,2020,1342
736,Tokushima-ken,2000,824
737,Tokushima-ken,2001,822
738,Tokushima-ken,2002,820
739,Tokushima-ken,2003,818
740,Tokushima-ken,2004,814
741,Tokushima-ken,2005,810
742,Tokushima-ken,2006,806
743,Tokushima-ken,2007,800
744,Tokushima-ken,2008,795
745,Tokushima-ken,2009,790
746,Tokushima-ken,2010,785
747,Tokushima-ken,2011,780
748,Tokushima-ken,2012,775
749,Tokushima-ken,2013,769
750,Tokushima-ken,2014,763
751,Tokushima-ken,2015,756
752,Tokushima-ken,2016,750
753,Tokushima-ken,2017,744
754,Tokushima-ken,2018,736
755,Tokushima-ken,2019,728
756,Tokushima-ken,2020,720
757,Kagawa-ken,2000,1023
758,Kagawa-ken,2001,1022
759,Kagawa-ken,2002,1020
760,Kagawa-ken,2003,1018
761,Kagawa-ken,2004,1016
762,Kagawa-ken,2005,1012
763,Kagawa-ken,2006,1009
764,Kagawa-ken,2007,1006
765,Kagawa-ken,2008,1003
766,Kagawa-ken,2009,999
767,Kagawa-ken,2010,996
768,Kagawa-ken,2011,992
769,Kagawa-ken,2012,989
770,Kagawa-ken,2013,985
771,Kagawa-ken,2014,980
772,Kagawa-ken,2015,976
773,Kagawa-ken,2016,973
774,Kagawa-ken,2017,968
775,Kagawa-ken,2018,963
776,Kagawa-ken,2019,958
777,Kagawa-ken,2020,950
778,Ehime-ken,2000,1493
779,Ehime-ken,2001,1490
780,Ehime-ken,2002,1485
781,Ehime-ken,2003,1481
782,Ehime-ken,2004,1475
783,Ehime-ken,2005,1468
784,Ehime-ken,2006,1460
785,Ehime-ken,2007,1453
786,Ehime-ken,2008,1445
787,Ehime-ken,2009,1438
788,Ehime-ken,2010,1431
789,Ehime-ken,2011,1424
790,Ehime-ken,2012,1415
791,Ehime-ken,2013,1406
792,Ehime-ken,2014,1396
793,Ehime-ken,2015,1385
794,Ehime-ken,2016,1377
795,Ehime-ken,2017,1368
796,Ehime-ken,2018,1357
797,Ehime-ken,2019,1346
798,Ehime-ken,2020,1335
799,Kochi-ken,2000,814
800,Kochi-ken,2001,812
801,Kochi-ken,2002,810
802,Kochi-ken,2003,806
803,Kochi-ken,2004,802
804,Kochi-ken,2005,796
805,Kochi-ken,2006,790
806,Kochi-ken,2007,783
807,Kochi-ken,2008,775
808,Kochi-ken,2009,769
809,Kochi-ken,2010,764
810,Kochi-ken,2011,758
811,Kochi-ken,2012,751
812,Kochi-ken,2013,743
813,Kochi-ken,2014,736
814,Kochi-ken,2015,728
815,Kochi-ken,2016,721
816,Kochi-ken,2017,714
817,Kochi-ken,2018,707
818,Kochi-ken,2019,699
819,Kochi-ken,2020,692
820,Fukuoka-ken,2000,5016
821,Fukuoka-ken,2001,5030
822,Fukuoka-ken,2002,5039
823,Fukuoka-ken,2003,5045
824,Fukuoka-ken,2004,5050
825,Fukuoka-ken,2005,5050
826,Fukuoka-ken,2006,5057
827,Fukuoka-ken,2007,5061
828,Fukuoka-ken,2008,5062
829,Fukuoka-ken,2009,5064
830,Fukuoka-ken,2010,5072
831,Fukuoka-ken,2011,5081
832,Fukuoka-ken,2012,5089
833,Fukuoka-ken,2013,5096
834,Fukuoka-ken,2014,5099
835,Fukuoka-ken,2015,5102
836,Fukuoka-ken,2016,5113
837,Fukuoka-ken,2017,5123
838,Fukuoka-ken,2018,5131
839,Fukuoka-ken,2019,5134
840,Fukuoka-ken,2020,5135
841,Saga-ken,2000,877
842,Saga-ken,2001,876
843,Saga-ken,2002,874
844,Saga-ken,2003,872
845,Saga-ken,2004,870
846,Saga-ken,2005,866
847,Saga-ken,2006,863
848,Saga-ken,2007,860
849,Saga-ken,2008,856
850,Saga-ken,2009,853
851,Saga-ken,2010,850
852,Saga-ken,2011,847
853,Saga-ken,2012,845
854,Saga-ken,2013,841
855,Saga-ken,2014,837
856,Saga-ken,2015,833
857,Saga-ken,2016,829
858,Saga-ken,2017,825
859,Saga-ken,2018,821
860,Saga-ken,2019,817
861,Saga-ken,2020,811
862,Nagasaki-ken,2000,1517
863,Nagasaki-ken,2001,1512
864,Nagasaki-ken,2002,1505
865,Nagasaki-ken,2003,1498
866,Nagasaki-ken,2004,1490
867,Nagasaki-ken,2005,1479
868,Nagasaki-ken,2006,1467
869,Nagasaki-ken,2007,1455
870,Nagasaki-ken,2008,1443
871,Nagasaki-ken,2009,1434
872,Nagasaki-ken,2010,1427
873,Nagasaki-ken,2011,1418
874,Nagasaki-ken,2012,1408
875,Nagasaki-ken,2013,1397
876,Nagasaki-ken,2014,1387
877,Nagasaki-ken,2015,1377
878,Nagasaki-ken,2016,1367
879,Nagasaki-ken,2017,1355
880,Nagasaki-ken,2018,1341
881,Nagasaki-ken,2019,1327
882,Nagasaki-ken,2020,1312
883,Kumamoto-ken,2000,1859
884,Kumamoto-ken,2001,1860
885,Kumamoto-ken,2002,1857
886,Kumamoto-ken,2003,1852
887,Kumamoto-ken,2004,1848
888,Kumamoto-ken,2005,1842
889,Kumamoto-ken,2006,1838
890,Kumamoto-ken,2007,1832
891,Kumamoto-ken,2008,1826
892,Kumamoto-ken,2009,1821
893,Kumamoto-ken,2010,1817
894,Kumamoto-ken,2011,1813
895,Kumamoto-ken,2012,1807
896,Kumamoto-ken,2013,1801
897,Kumamoto-ken,2014,1795
898,Kumamoto-ken,2015,1786
899,Kumamoto-ken,2016,1775
900,Kumamoto-ken,2017,1767
901,Kumamoto-ken,2018,1759
902,Kumamoto-ken,2019,1749
903,Kumamoto-ken,2020,1738
904,Oita-ken,2000,1221
905,Oita-ken,2001,1220
906,Oita-ken,2002,1219
907,Oita-ken,2003,1217
908,Oita-ken,2004,1214
909,Oita-ken,2005,1210
910,Oita-ken,2006,1207
911,Oita-ken,2007,1206
912,Oita-ken,2008,1204
913,Oita-ken,2009,1200
914,Oita-ken,2010,1197
915,Oita-ken,2011,1192
916,Oita-ken,2012,1186
917,Oita-ken,2013,1180
918,Oita-ken,2014,1173
919,Oita-ken,2015,1166
920,Oita-ken,2016,1160
921,Oita-ken,2017,1152
922,Oita-ken,2018,1143
923,Oita-ken,2019,1134
924,Oita-ken,2020,1124
925,Miyazaki-ken,2000,1170
926,Miyazaki-ken,2001,1168
927,Miyazaki-ken,2002,1165
928,Miyazaki-ken,2003,1162
929,Miyazaki-ken,2004,1159
930,Miyazaki-ken,2005,1153
931,Miyazaki-ken,2006,1150
932,Miyazaki-ken,2007,1146
933,Miyazaki-ken,2008,1141
934,Miyazaki-ken,2009,1138
935,Miyazaki-ken,2010,1135
936,Miyazaki-ken,2011,1130
937,Miyazaki-ken,2012,1125
938,Miyazaki-ken,2013,1119
939,Miyazaki-ken,2014,1112
940,Miyazaki-ken,2015,1104
941,Miyazaki-ken,2016,1097
942,Miyazaki-ken,2017,1091
943,Miyazaki-ken,2018,1084
944,Miyazaki-ken,2019,1077
945,Miyazaki-ken,2020,1070
946,Kagoshima-ken,2000,1786
947,Kagoshima-ken,2001,1782
948,Kagoshima-ken,2002,1776
949,Kagoshima-ken,2003,1770
950,Kagoshima-ken,2004,1763
951,Kagoshima-ken,2005,1753
952,Kagoshima-ken,2006,1744
953,Kagoshima-ken,2007,1733
954,Kagoshima-ken,2008,1721
955,Kagoshima-ken,2009,1712
956,Kagoshima-ken,2010,1706
957,Kagoshima-ken,2011,1697
958,Kagoshima-ken,2012,1687
959,Kagoshima-ken,2013,1675
960,Kagoshima-ken,2014,1662
961,Kagoshima-ken,2015,1648
962,Kagoshima-ken,2016,1637
963,Kagoshima-ken,2017,1626
964,Kagoshima-ken,2018,1614
965,Kagoshima-ken,2019,1602
966,Kagoshima-ken,2020,1588
967,Okinawa-ken,2000,1318
968,Okinawa-ken,2001,1327
969,Okinawa-ken,2002,1336
970,Okinawa-ken,2003,1345
971,Okinawa-ken,2004,1353
972,Okinawa-ken,2005,1362
973,Okinawa-ken,2006,1369
974,Okinawa-ken,2007,1374
975,Okinawa-ken,2008,1378
976,Okinawa-ken,2009,1385
977,Okinawa-ken,2010,1393
978,Okinawa-ken,2011,1402
979,Okinawa-ken,2012,1411
980,Okinawa-ken,2013,1419
981,Okinawa-ken,2014,1426
982,Okinawa-ken,2015,1434
983,Okinawa-ken,2016,1442
984,Okinawa-ken,2017,1448
985,Okinawa-ken,2018,1454
986,Okinawa-ken,2019,1462
987,Okinawa-ken,2020,1467

  2. PostgreSQL構築

    sudo apt install postgresql postgresql-contrib
sudo su - postgres
psql
  postgres=> CREATE ROLE test WITH LOGIN PASSWORD 'test';
  postgres=> GRANT pg_read_server_files TO test;
  postgres=> \du
  postgres=> CREATE DATABASE test WITH OWNER test;
  postgres=> \l
  postgres=> exit;
exit
sudo vi /etc/postgresql/14/main/pg_hba.conf
  local host all all scram-sha-256
sudo vi /etc/postgresql/14/main/postgresql.conf
  search_path = '"$user", $user, public'  # schema names
sudo systemctl restart postgresql
psql -d test -U test
  test=> CREATE SCHEMA test;
  test=> \dn
  test=> CREATE TABLE test("ID" int, "prefecture" varchar(32), "year" int, "population" int);
  test=> \COPY test FROM './population_2000-2020.csv' WITH CSV HEADER;

  3. uvインストール

    curl -LsSf https://astral.sh/uv/install.sh | sh

    ※ uvはPythonの管理ツール(https://docs.astral.sh/uv/)

  4. 仮想環境作成

    cd <任意のディレクトリ>
uv init mcp_test --python 3.12
cd mcp_test
uv sync
. .venv/bin/activate

  5. ディレクトリ作成

    cd mcp_test
mkdir -p agent_sample/agent
mkdir server_sample

  6. ADKインストール

    cd mcp_test
uv add "google-adk"
uv add "litellm"

    ※ litellmは、google-adkにGoogle外のLLM（OpenAI等）を渡すため
    https://docs.litellm.ai/docs/providers/

  7. MCP Python SDKインストール

    cd mcp_test
uv add "mcp[cli]"

  8. pyodbcインストール

    cd mcp_test
uv add "pyodbc"

  9. ODBCドライバインストール

    sudo apt install odbc-postgresql
sudo odbcinst -i -d -f /etc/odbcinst.ini
odbcinst -q -d

  10. VSCodeの仮想環境認識設定

    mcp_test/.vscode/settings.json

    "python.venvFolders": [
  ".venv"
]

  11. Agent実装例

    mcp_test/agent_samples/agent.py 
    from contextlib import AsyncExitStack
from dotenv import load_dotenv
from google.adk.agents import LlmAgent
from google.adk.models.lite_llm import LiteLlm
from google.adk.tools.mcp_tool.mcp_toolset import MCPToolset, SseServerParams


# load API key
load_dotenv(".env")


async def create_agent():
    async_exit_stack = AsyncExitStack()
    tools, _ = await MCPToolset.from_server(
        connection_params=SseServerParams(url="http://localhost:8080/sse"),
        async_exit_stack=async_exit_stack,
    )
    agent = LlmAgent(
        # model=LiteLlm(model="openrouter/anthropic/claude-3.7-sonnet:thinking"),
        model=LiteLlm(model="openai/gpt-4o"),
        name="agent",
        instruction="あなたは有能なアシスタントです。日本語で回答します。"
        "適切なデータがないかデータベースを確認します。データベースにはMCPでアクセスします。"
        "最初に、MCPサーバでテーブル一覧を取得して、質問に必要なデータがないか見てみましょう。",
        tools=[
            *tools,
        ],
    )
    return agent, async_exit_stack


    root_agent = create_agent()

    ※ 最小限の実装。root_agentにLiteLlmを通してOpenrouterOpenAIのAgentを渡す
    ※ .envに~~OPENROUTER_API_KEY=<あなたのOpenrouterのキー>~~`OPENAI_API_KEY=<あなたのOpenAIのキー>を記載し、load_dotenvで読み込んでおく ※ adk webを実行後、http://127.0.0.1:8000`を開くとWebUIが出る

  12. MCP Server実装例

    mcp_test/server_samples/server.py 
    import os
import io
import csv
import click
from contextlib import asynccontextmanager
from collections.abc import AsyncIterator
from dataclasses import dataclass
from mcp.server.fastmcp import Context, FastMCP
import pyodbc


class PostgreDB:
    def __init__(
        self,
        database: str,
        user: str,
        password: str,
        server: str = "127.0.0.1",
        port: str = "5432",
    ):
        self.database = database
        self.user = user
        self.password = password
        self.server = server
        self.port = port

    def __await__(self):
        return self.__async_init().__await__()

    async def __async_init(self):
        self.__con: pyodbc.Connection = await self.__connect()
        self.__schemas: list = await self.__get_schemas()
        self.__tables: list[dict] = await self.__get_tables()
        return self

    async def __connect(self) -> pyodbc.Connection:
        os.environ["PGUSER"] = self.user
        os.environ["PGPASSWORD"] = self.password
        connection_string = f"DRIVER={{PostgreSQL Unicode}}; SERVER={self.server}; PORT={self.port}; DATABASE={self.database};"
        return pyodbc.connect(connection_string)

    async def __get_schemas(self) -> list[dict]:
        con: pyodbc.Connection = self.con
        sql = "SELECT schema_name FROM information_schema.schemata;"
        result = con.execute(sql).fetchall()
        schemas = list(
            set(self.rows2list(result)) - {"pg_catalog", "information_schema"}
        )
        return schemas

    async def __get_table(self, schema: str) -> list[str]:
        con: pyodbc.Connection = self.con
        sql = f"SELECT tablename FROM pg_tables where schemaname = '{schema}';"
        result = con.execute(sql).fetchall()
        tables = self.rows2list(result)
        return tables

    async def __get_tables(self) -> list[dict]:
        schemas: list = self.schemas if self.schemas else self.__get_schemas
        result = {}
        for schema in schemas:
            tables = await self.__get_table(schema)
            result[schema] = tables
        return result

    def rows2dict(self, rows: list[pyodbc.Cursor]) -> list[dict]:
        if len(rows) == 0:
            return []
        keys = [k[0] if len(k) > 0 else k for k in rows[0].cursor_description]
        result = [dict(zip(keys, row)) for row in rows]
        return result

    def rows2list(self, rows: list[pyodbc.Cursor]) -> list[str]:
        if len(rows) == 0:
            return []
        result = [str(row[0]) for row in rows]
        return result

    def rows2csv(self, rows: list[pyodbc.Cursor]) -> str:
        if len(rows) == 0:
            return ""
        d = self.rows2dict(rows)
        buffer = io.StringIO()
        writer = csv.DictWriter(buffer, fieldnames=d[0].keys(), quoting=csv.QUOTE_ALL)
        writer.writeheader()
        writer.writerows(d)
        result = buffer.getvalue()
        return result

    async def select(self, schema: str, table: str, limit: int) -> list[dict]:
        sql = f"SELECT * FROM {schema}.{table}"
        if limit > 0:
            sql += f" limit {limit}"
        sql += ";"
        result = self.con.execute(sql).fetchall()
        return self.rows2csv(result)

    @property
    def con(self):
        return self.__con

    @property
    def schemas(self):
        return self.__schemas

    @property
    def tables(self):
        return self.__tables


@dataclass
class AppContext:
    db: PostgreDB


@asynccontextmanager
async def app_lifespan(servver: FastMCP) -> AsyncIterator[AppContext]:
    # since this is sample, hard-coded.
    db: PostgreDB = await PostgreDB(
        database="test",
        user="test",
        password="test",
    )
    try:
        yield AppContext(db=db)
    finally:
        db.con.close()


# Create an MCP server
mcp = FastMCP("posetgresql_test", lifespan=app_lifespan, host="127.0.0.1", port="8080")


# Resources
# Note: ADK Web does not recognize the resources (maybe an oversight), so I use them as a tools.
#
# @mcp.resource(uri="schemas://main", mime_type="application/json")
# async def get_schemas() -> list[str]:
#    """This returns schemas.
#
#    Returns:
#        list[str]: [schema1, schema2, ...]
#    """
#    ctx = mcp.get_context()
#    db: PostgreDB = ctx.request_context.lifespan_context.db
#    return db.schemas
#
#
# @mcp.resource(uri="schemas://{schema}/tables", mime_type="application/json")
# async def get_tables(schema: str) -> list[str]:
#    """This returns tables in the schema.
#
#    Args:
#        schema (str): Specify the schema.
#
#    Returns:
#        list[str]: [table1, table2, ...]
#    """
#    ctx = mcp.get_context()
#    db: PostgreDB = ctx.request_context.lifespan_context.db
#    return db.tables[schema]
#
#
# @mcp.resource(uri="tables://main", mime_type="application/json")
# async def get_all_tables() -> dict[list[str]]:
#    """This returns all tables in all schemas.
#
#    Returns:
#        dict[list[str]]:  {schema1: [table1, table2, ...], schdma2: [], ...}.
#    """
#    ctx = mcp.get_context()
#    db: PostgreDB = ctx.request_context.lifespan_context.db
#    return db.tables


# Tools
@mcp.tool()
async def get_all_tables(ctx: Context) -> dict[list[str]]:
    """This returns all tables in all schemas.

    Returns:
        dict[list[str]]:  {schema1: [table1, table2, ...], schdma2: [], ...}.
    """
    db: PostgreDB = ctx.request_context.lifespan_context.db
    return db.tables


@mcp.tool()
async def select(ctx: Context, schema: str, table: str, limit: int) -> list[dict]:
    """Return the result executed a SELECT claude.

    Args:
        schema (str): Specify the shema.
        table (str): Specify the table.
        limit (int): Specify The number of limit. 0 means no use of limit.

    Returns:
        list[dict]: [{"col1": val1, "col2": val2, ...}, ...]
    """
    db: PostgreDB = ctx.request_context.lifespan_context.db
    return await db.select(schema, table, limit)


@click.command()
@click.option(
    "--transport",
    "-t",
    type=str,
    default="sse",
    help='Specify "sse" or "stdio". Defaults to "sse"',
)
def main(transport: str):
    if transport not in ["sse", "stdio"]:
        raise click.BadParameter(f'Transport should be "sse" or "stdio". {transport=}')
    mcp.run(transport=transport)


if __name__ == "__main__":
    main()

    ※ ADKだとresourcesが参照されてなさそうだったので（おそらく設定もれだが）、テーブル一覧等もtoolsに実装。

起動手順

  1. MCP Server起動

    1. MCP Inspector起動

      npx @modelcontextprotocol/inspector

      ※ MCP Inspector起動は、mcp cliによる以下コマンドの延長でも実行可能（MCPサーバは、STDIOモードで起動される）

      mcp dev server.py

    2. サーバ起動

      • ダイレクトに起動

        以下のスクリプトがserver.pyに記載されている必要がある。

        if __name__ == "__main__":
    mcp.run(transport="sse")

      • ポートは8080にする

        MCPクライアントがデフォルトで8000を使うので避ける。

        mcp = FastMCP("posetgresql_test", lifespan=app_lifespan, host="127.0.0.1", port="8080")

      起動コマンドは以下。

      python server.py

      または

      uv run server.py

    3. MCP Inspectorから動作確認

      Transport Type: SSE
      URL: http://localhost:8080/sse

      image.png

  2. MCP Client起動

    1. MCP Client起動

      adk web

      image.png

    ※ システムプロンプトにMCPサーバを確認するよう入力してある

0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
Sign upLogin
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?