accurate-schema-mcp
Schema-aware MCP server for Accurate Online, providing tools to list resources, lookup endpoint schemas, make authenticated API calls, and refresh database host resolution.
README
accurate-schema-mcp
Schema-aware Model Context Protocol server for Accurate Online.
Features
- Always-fresh OpenAPI spec. Refetched on every server startup (i.e. each time Claude Desktop launches) and written over the cached copy, so new endpoints appear with no code change and no manual step. If the download fails, the cached copy is used.
- Host auto-resolution. Nothing is hardcoded — the server asks Accurate
which host your API Token's database lives on.
refresh_hostre-resolves if the database migrates mid-session. - Delete kill switch.
generic_callrefuses delete endpoints (HTTPDELETEand*/delete.do,*/bulk-delete.do) unless you explicitly opt in. - Verified list fields. Accurate's spec documents no response schema for
list.do, so real field names are shipped inlist_fields.jsonand merged into every<resource>/listlookup. - Python 3.10+, zero config files beyond
.env, onlymcp+requestsas dependencies.
Requirements
- Python version >= 3.10
uvfor dependency management- An Accurate Online API Token — steps on how to get
Bearer TokenandAPI Secretfrom https://drive.google.com/file/d/1_EKt6DE6U-6vkmv4M_vizmC-27wXjXa0/view.
Install
git clone https://github.com/aol-integration/accurate-schema-mcp.git
cd accurate-schema-mcp
uv sync # creates .venv from the exact versions pinned in uv.lock
Configure
cp .env.example .env
Edit .env:
ACCURATE_BEARER_TOKEN=<your_api_token>
ACCURATE_API_SECRET=<your_api_token_secret_key>
# generic_call refuses delete endpoints unless this is true.
ACCURATE_ENABLE_DELETE=false
See this page for more information on Accurate Online API Token authentication.
Add to Claude Desktop
-
Open Claude Desktop.
-
Go to Settings → Developer → Edit Config. This opens the folder containing
claude_desktop_config.json. -
Open
claude_desktop_config.jsonin a text editor. -
Paste this in. If the file already has
mcpServers, add only the"accurate-schema"entry inside it.{ "mcpServers": { "accurate-schema": { "command": "uv", "args": [ "run", "--directory", "/absolute/path/to/accurate-schema-mcp", "python", "-m", "accurate_schema_mcp.server" ] } } } -
Replace
/absolute/path/to/accurate-schema-mcpwith your clone's real path (pwdinside the folder).--directoryalso sets the working directory, so.envis picked up. -
Save the file and fully quit Claude Desktop (Cmd+Q on macOS), then reopen it. The config is only read at launch.
-
Check the tools menu in the chat input —
accurate-schemashould appear with 4 tools.
The server speaks MCP over stdio.
Tools (4)
| Tool | Arguments | Description |
|---|---|---|
list_resources |
— | All resource names (item, vendor, purchase-invoice, sales-order, …). Call first if you don't know the resource. |
schema_lookup |
endpoint |
Fields, query params and required-ness for one endpoint. Accepts '<resource>/<action>' (e.g. 'purchase-invoice/save') or just 'item' to list that resource's endpoints. |
generic_call |
endpoint, params?, body? |
Call any endpoint. params for GET/DELETE, body for POST (save, bulk-save). |
refresh_host |
— | Re-resolve which host the API Token's database lives on. |
Typical flow
# 1) Which resources exist?
list_resources()
# -> ["access-privilege", "branch", "customer", "item", "purchase-invoice", ...]
# 2) What does this endpoint want?
schema_lookup(endpoint="purchase-invoice/save")
# -> {"method": "POST", "path": "/api/purchase-invoice/save.do",
# "body": {"vendorNo": {"required": true, ...}, ...}}
# 3) Call it
generic_call(
endpoint="item/list",
params={"fields": "id,name,unitPrice", "sp.pageSize": 50},
)
Authentication
Each request is signed fresh — no session or token-refresh step:
X-Api-Timestamp: <unix epoch milliseconds>
X-Api-Signature: HMAC-SHA256(api_secret, timestamp)
Authorization: Bearer <bearer_token>
The host is resolved at startup by calling
https://account.accurate.id/api/api-token.do with the same headers. The host
comes straight from d.database.host in that response (e.g.
https://public.accurate.id) — one field, no payload walking. A 401 means the
token or signature is invalid; note
that Accurate also signals failure with HTTP 200 + {"s": false}, which is
handled as an auth error.
GET requests retry up to 3 times with backoff on 502/503/504. Mutating
requests are never retried — they may already have succeeded.
Schema cache
The OpenAPI spec is downloaded from
https://account.accurate.id/open-api/json.do into
schema/accurate_openapi.json on every server startup, overwriting the
previous copy (first run simply creates it). The write is atomic — the spec is
staged to a .tmp file and swapped in — and a response with no paths is
rejected, so a bad download can't clobber a good cache. If the fetch fails and
a cached copy exists, the server logs a warning and starts with the cache.
To refresh it by hand:
uv run python -m accurate_schema_mcp.fetch_schema --force
Environment variables
| Variable | Required | Notes |
|---|---|---|
ACCURATE_BEARER_TOKEN |
yes | Accurate Online → Accurate Store → API Token |
ACCURATE_API_SECRET |
yes | Accurate Online → Developer Area |
ACCURATE_ENABLE_DELETE |
no | false by default; generic_call refuses delete endpoints until this is true |
Project layout
accurate_schema_mcp/
├── server.py MCP entrypoint (stdio)
├── tools_schema.py the four tools + delete guard
├── client.py HMAC auth, host resolution, HTTP
├── config.py env vars / .env loader (no dependency)
├── schema_index.py parses the OpenAPI spec into a flat index
├── fetch_schema.py downloads the spec (refreshed every startup)
└── list_fields.json verified list.do response fields (absent from the spec)
schema/
└── accurate_openapi.json cached spec
Verifying it works
uv run python -c "from accurate_schema_mcp.client import get_client; c = get_client(); print(c.base_url)"
Printing a base URL like https://public.accurate.id/accurate confirms the
credentials, the signature and host resolution all work. A 401 usually means
a wrong secret, a wrong bearer token, or whitespace pasted into .env.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。