SignalEDI MCP Server
Enables AI assistants to parse, validate, send, and inspect X12 EDI documents via the SignalEDI Core API, with optional QuickBooks integration and a demo mode for keyless local testing.
README
SignalEDI MCP Server
A Model Context Protocol server that exposes the
SignalEDI Core API (/api/v1) as tools, so AI clients — Claude Desktop, Cursor,
Windsurf, Claude Code, and any other MCP host — can parse, validate, send, and
inspect X12 EDI documents directly in a conversation or agent loop.
It is mostly a thin adapter over the hosted SignalEDI engine, with a few local X12 helper surfaces for demo mode and model grounding: synthetic templates, a small validation/error dictionary, prompts, and bundled resources. The hosted API remains the source of truth for production parse, validate, send, transaction, kit, and QuickBooks flows.
| Package | @signaledi/mcp-server (npm) |
| MCP Registry | io.github.SignalEDI/mcp-server (registry.modelcontextprotocol.io) |
| Transport | stdio |
| Runtime | Node 18+ (uses global fetch) |
| Runtime deps | @modelcontextprotocol/sdk only |
| Auth | Optional — demo mode without a key (parse/validate via public playground; generate/dictionary tools local); set SIGNALEDI_API_KEY for send/partner/transaction/QBO tools. |
Discoverable in the official MCP Registry and the directories that crawl it
(Smithery, Glama, mcp.so, PulseMCP). The server.json in this folder is the registry
manifest; package.json carries the matching mcpName ownership marker.
Tools
| Tool | What it does |
|---|---|
parse_edi |
Parse a raw X12 interchange into structured JSON + a validation summary. |
validate_edi |
Validate a raw interchange against X12 structural rules (summary only). |
send_outbound_document |
Serialize a JSON payload to EDI and send it to a trading partner (async, webhook-acked). |
list_transactions |
List your recent transactions (newest first), scoped to the API key. |
get_transaction |
Fetch one transaction by id with full lifecycle status. |
quickbooks_status |
Show QuickBooks Online connection status without returning tokens. |
quickbooks_sync_to_qbo |
Push eligible EDI transactions into QuickBooks Online. |
quickbooks_export_to_edi |
Pull QBO invoices or purchase orders and emit outbound EDI payloads. |
quickbooks_list_entities |
List QBO invoices, purchase orders, customers, vendors, or items for preview/mapping. |
quickbooks_disconnect |
Revoke and remove the workspace QBO connection. |
list_partner_kits |
List packaged API kits (requires key). |
get_partner_kit |
Fetch one kit by kitId, with partnerId accepted as an alias (requires key). |
explain_edi_error |
Local X12 dictionary lookup for ack/validation errors. |
generate_test_document |
Render synthetic X12 850/810/856/837 samples (works in demo mode). |
lookup_x12 |
Search segment and acknowledgement reference. |
validate_x12_structure |
Alias for validate_edi, tuned for tool discovery. |
parse_segments |
Alias for parse_edi, tuned for tool discovery. |
lookup_element_definition |
Alias for lookup_x12; accepts query or element. |
Bad arguments and API errors are returned as structured MCP tool errors with stable
codes (so the model can recover from them) rather than crashing the server. Results
include both backwards-compatible text content and native structuredContent.
Mutation safety
Tools that send EDI, write to QuickBooks, or disconnect QuickBooks require both
confirm: true and a unique idempotencyKey (at least eight characters). The
server adds a request id, tool name, and idempotency header to outbound API calls.
Dry-run QuickBooks export remains available without confirmation. Tool listings
advertise read-only versus destructive behavior and required capability scopes.
The client rejects non-HTTPS base URLs except for localhost, caps EDI payloads at
64 KiB, retries transient failures within a total deadline, and emits redacted
per-tool latency/success metrics to stderr. Set SIGNALEDI_MCP_TELEMETRY=0 to
disable local metric lines.
Quick start
- Try without a key (demo mode). Run
npx -y @signaledi/mcp-serverwith no env vars —parse_ediandvalidate_edicall the public playground;generate_test_document,explain_edi_error,lookup_x12, and the alias tools use local X12 templates/dictionary data. Keyed tools return a structureddemo_modeerror with a link to create a key. - Add a platform key for production flows. Create a workspace key with the
platformscope at signaledi.com/console/keys and setSIGNALEDI_API_KEY.
One-click install
- Cursor: Install in Cursor (same deeplink as the developer site chip).
- Claude Code:
claude mcp add signaledi -- npx -y @signaledi/mcp-server
Claude Desktop / Claude Code
Add to your MCP config (claude_desktop_config.json, or .mcp.json for Claude Code):
{
"mcpServers": {
"signaledi": {
"command": "npx",
"args": ["-y", "@signaledi/mcp-server"],
"env": {
"SIGNALEDI_API_KEY": "sk_live_…"
}
}
}
}
Cursor / Windsurf
Same shape under the editor's MCP settings (command: npx, args: ["-y", "@signaledi/mcp-server"], and the SIGNALEDI_API_KEY env var).
Demo mode
Without a key, the server runs in demo mode (stderr notice): parse_edi and validate_edi call the public playground; local template/dictionary tools run without network; keyed tools return JSON { "error": "demo_mode" }; successful demo results append — demo mode; responses rate-limited.
Configuration
| Env var | Required | Default | Notes |
|---|---|---|---|
SIGNALEDI_API_KEY |
— (demo mode) / ✅ (full access) | — | Workspace key with the platform scope for send/partner/QBO tools. |
SIGNALEDI_BASE_URL |
— | https://signaledi.com |
Point at a custom domain or a preview deployment. |
The server speaks JSON-RPC over stdout; all logs go to stderr, so it never corrupts the protocol channel.
Example prompts
"Parse this 850 and tell me the PO number and ship-to." (paste raw EDI) "Validate this interchange and list any structural errors." "Show my last 10 transactions and flag any that missed SLA." "Send an 810 invoice to partner
acme-cowith these line items…"
Local development
node test.mjs # dependency-free unit tests (mock fetch, no network)
SIGNALEDI_API_KEY=… npm start # run the stdio server against the live API
The client (src/client.mjs) and tools (src/tools.mjs) are pure and unit-tested;
src/index.mjs is the only file that depends on the MCP SDK.
How it fits
MCP client (Claude/Cursor/…) ⇄ @signaledi/mcp-server (stdio) ⇄ SignalEDI Core API (/api/v1)
└ local X12 templates/dictionary/resources for demo + guidance
See the SDK for a programmatic TypeScript client and
docs/openapi/v1 for the full API reference.
GitHub mirror
The public GitHub repo SignalEDI/mcp-server mirrors this folder only. The GitLab monorepo stays the source of truth; the mirror updates on each npm publish (see MIRROR.md and the operator runbook docs/internal/runbooks/GITHUB_MCP_MIRROR.md).
Examples
| Script | Purpose |
|---|---|
examples/parse-demo.mjs |
Parse synthetic 850 via demo-mode client |
examples/generate-validate.mjs |
Generate 850 locally, then validate |
examples/agent-transcript.md |
Illustrative agent conversation (no secrets) |
Run from this directory: node examples/parse-demo.mjs
Issue templates for the GitHub repo live under examples/.github/ISSUE_TEMPLATE/ (copied to repo root on mirror sync). Layout: examples/github-mirror-layout.md.
Mirror security checklist
Before every mirror push:
- No internal URLs, API keys, engine code, or employee/customer data in the tree
- Examples use synthetic EDI only; transcripts are fictional
- Badges and links point at public npm, MCP registry, and signaledi.com only
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。