
Vendor Statement Open Items
io.github.theluckystrike由 GitHub 账号 theluckystrike 发布v0.25.0更新于 Oct 26, 2026
Compare supplied vendor statement and AP open items locally; keep missing and ambiguous evidence.
概览
本地 MCP 服务器,将供应商对账单与应付未清项导出进行比对,报告缺失、冲突和无法配对的条目。
- 可以做什么
- 按指定截止日期比对两份提供的未清项清单,输出已匹配、仅在对账单、仅在账簿、差异和引用歧义等分组
- 列出、读取和删除已保存的报告,并分页查看对账单行、账簿行、分组及分组成员
- 将报告导出为 CSV,每个原始来源记录一行
- 查询离线许可证等级并激活 Pro 许可证密钥
- 适用场景
- 当你已把供应商对账单和应付未清项导出整理为结构化数据,需要一种确定性的、保留全部原始行的比对并解释每项发现时使用。不适用于余额核对、付款匹配、连接 ERP、读取 PDF 或审批付款。
- 运行要求
- 需要 Node.js 18 或更高版本及 MCP 客户端;Claude Desktop 可直接打开 .mcpb 包,其他客户端需从源码构建(npm 包未发布)。免费版无需账号或 API 密钥。可选 MCP_LICENSE_KEY 用于离线 Pro。数据保存在本地 XDG 数据目录下。
安装
在 SourceWeft 中
- 打开 控制台中的 Vendor Statement Open Items,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
Supplier statement reconciliation MCP server
Compare a supplied supplier statement with an accounts-payable open-item export. Find invoices or credit notes present on only one side, conflicting outstanding amounts or dates, and repeated references that cannot be paired safely. This local MCP server preserves every original row and explains each finding. It checks two supplied open-item lists; it does not reconcile opening/closing balances, match payments, connect to an ERP, read PDFs, or approve payment.
Built by theluckystrike. MIT. Support: [email protected].
Try it before installing
Run the free supplier statement example in your browser. Use the synthetic example to inspect missing items and amount/date discrepancies. The browser proof and this server use the same comparison core. No account or API key is needed for the local server's free tier. Product and installation options.
These comparisons require the relevant source data. They do not retrieve missing documents or authorize a payment. Full diagnostics and CSV remain free; the paid upgrade removes only the limit on saved report count.
Install
Requires Node.js 18 or newer and an MCP client. Claude Desktop can open the
supplier-statement-check.mcpb v0.25.0 package.
Other clients can build the source below. The npm package is not published;
npx @theluckystrike/mcp-supplier-statement-check is not an installation route.
These commands build only the shared license dependency and this server. Replace
/absolute/path below with the source checkout location. If your MCP client cannot
find node, use its full executable path.
Claude Desktop: merge into mcpServers in
~/Library/Application Support/Claude/claude_desktop_config.json on macOS, or
%APPDATA%\Claude\claude_desktop_config.json on Windows:
Claude Code:
Cursor: merge into .cursor/mcp.json in the project or ~/.cursor/mcp.json:
VS Code: put the entry under servers in .vscode/mcp.json:
Windsurf's legacy Cascade agent: use the Claude Desktop mcpServers entry in
~/.codeium/windsurf/mcp_config.json. That file applies to Cascade; a different
agent can use a different configuration route.
Cline: open MCP Servers, Configure, Configure MCP Servers, and add:
These are local stdio configurations. This server has no hosted endpoint or browser
chat connector. Reconnect your client, call license_status, then run the example
below. Tool discovery alone does not verify a comparison.
Tools
A missing item and a conflicting amount
Call supplier_statement_check with these synthetic inputs:
The result has four groups: one match, one discrepancy with both amount_mismatch
and date_mismatch, one item only in the statement, and one only in the ledger.
Every supplied row remains available. The credit note's amount is positive and is
not netted against invoices. No balance, loss or recoverable amount is calculated.
If you add another INV-001 within either side's exact scope, its entire group
becomes ambiguous_reference; the tool does not choose the convenient amount pair.
Exact input contract and comparison rules
Supply a real YYYY-MM-DD as_of date and two arrays, each with 0–1000 records.
At least one side must contain a record. The date declares the intended common
snapshot; the tool cannot verify export freshness and does not filter rows by date.
Prepare both lists with consistent open-item scope before comparing them.
Each record requires record_id, entity_id, supplier_id, uppercase three-letter
currency, document_type (invoice or credit_note), document_reference, real
ISO document_date, and positive safe-integer outstanding_minor. IDs/references
are 1–160 characters without surrounding whitespace or control characters. IDs
must be unique within a side; the two sides may use the same ID. Unknown fields,
impossible dates, year 0000, zero/negative/decimal/unsafe amounts and over-limit
arrays are rejected. Currency syntax is checked without an external currency list.
Amounts are currency minor units: USD 12500 means $125.00. Credit notes use positive
absolute outstanding amounts and compare only with credit notes. Neither side is
required to use a common supplier name, but its exact supplier_id must agree.
Different entities, suppliers, currencies and document types remain separate.
The join key is exact (entity_id, supplier_id, currency, document_type, document_reference). Leading zeros, punctuation, case, accents, Unicode forms and
lookalike characters remain significant. No supplier inference, reference cleanup,
greedy amount matching, payment allocation or currency conversion occurs.
Missing rows and disagreements can result from export timing or scope. Repeated
references can be legitimate. Even complete agreement between supplied lists is not
proof that the supplier balance is correct: two incomplete exports can agree. The
status no_discrepancies_in_supplied_open_items makes only that limited observation.
Saved reports never become hidden cross-call history; include both lists every time.
Groups and rows are deterministically sorted. Each source row belongs to exactly one group; the engine does not expand every possible record pair. A group refers to source IDs qualified by their statement/ledger side, so identical IDs across sides cannot overwrite evidence. No network or telemetry requests are made.
Free evidence, previews and CSV
Every check returns retrieval.report_id. Unsaved previews use preview:<UUID> and
remain in memory for 30 minutes for the last two checks in this process. A third
check, restart or expiry may evict a preview; resubmit the free check. Saved UUIDs
persist on disk. Preview data is not written to a browser or filesystem.
report_page supports statement_records, ledger_records, groups, and
group_members. Group pages contain summary metadata and side-specific member
counts. Read complete source members with section:"group_members" and the exact
group_id. Follow next_offset until null. Pages and CSV remain free even after
the saved-report cap is reached.
Responses use a 512 KiB escaped MCP text-envelope budget. Larger reports return a
paginated overview; pages can contain fewer rows than requested to fit that budget.
CSV uses one row per original source record with its side, original fields, group
and reasons. Formula-like text is escaped; JSON source values are unchanged. Export
returns CSV text, never writes to a supplied filename, and repeats the header on
each page. Retain one header when joining pages. Follow actual returned and
next_offset values rather than assuming the requested count was returned.
Free and Pro
save:true at the free cap returns the diagnostic and free preview with
saved:false, save_error:"free_saved_report_cap", and an upgrade link. Delete a
saved report to reuse a slot. Storage failures also preserve the diagnostic.
Get Pro: $19 one time.
All servers: $39 bundle. Set MCP_LICENSE_KEY or
call license_activate; keys verify offline. Pro removes only the saved-report
count cap. It adds no accounting data, accuracy guarantee, ERP access or safety
approval. The saved-history upgrade is a commercial hypothesis, not proven ROI.
All data stays local. Reports live under
${XDG_DATA_HOME:-~/.local/share}/mcp-servers/supplier-statement-check/.
XDG_DATA_HOME must be absolute with real directory ancestors. Private JSON files
are written atomically with cross-process mutation locks. Reads validate original
inputs and recompute evidence rather than trusting edited results. License
activation writes the shared local license configuration. Manage accounting-data
access, retention and backups yourself.
Development
Build Docker from the monorepo root with
docker build -f servers/supplier-statement-check/Dockerfile .. The standalone
mirror receives its own self-contained Dockerfile during mirror generation.
Tests cover the comparison oracle, strict inputs, scope/Unicode isolation, complete
source coverage, ambiguity, large free pages/CSV, safe storage, concurrent save
caps and actual stdio calls. Signed-license tests use the existing local private
signer when available and skip if it is absent. Private signing keys are not
packaged or required by users.
来源:servers/supplier-statement-check/README.md,提交 3b2e499
工具
0版本历史
1- v0.25.0最新Oct 26, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分