Keeps invoices, test specs and design docs as schema-checked Markdown and TSV that an AI agent can edit and commit locally.
An invoice in a Word file is easy for a person to read and awkward for an agent to change safely. md-business inverts that. Documents live as Markdown with YAML frontmatter, each one declares a JSON Schema, Git keeps the history, and viewers turn the result into A4 PDF or a browser preview for whoever needs to approve it. The md-business MCP server is how an AI agent operates on those files.
| Type | Schema |
|---|---|
| Invoice, quote, receipt | invoice/v1 |
| Test specification | test-spec/v1 (and test-spec-tsv/v1 for sheets) |
| Design document | spec/v1 |
| API specification | api-spec/v1 |
| Database design | db-spec/v1, nosql-db-spec/v1 |
| Investigation report | investigation/v1 |
The invoice templates lean towards Japanese requirements, including qualified-invoice registration numbers and mixed 10% and 8% tax rates, with an English baseline as well. All templates use invented data.
The server runs inside the desktop app for Windows or macOS and starts when the app launches. Open a folder, go to the MCP tab, and either write a .mcp.json into the folder, which Claude Code picks up automatically, or copy a ready-made client entry. The entry is a local HTTP URL with a port and a bearer token; both persist across restarts. In a Git repository the app adds .mcp.json to .gitignore, because the token is a credential. A standalone server without the app also exists.
git_status, git_diff and git_commit in a repository. There is deliberately no pushThe md-business MCP server reaches no further than the folder open in the app, and the server listens on loopback only. It never receives API keys. Some channels, such as a PWA and a GitHub Action, are still planned, and database design schemas have no blank template yet.
If your agent also needs plain repository access, pair it with Git or Filesystem, or browse other developer tools.
Check the linked publisher documentation and your connected client's tool list. Names and availability can vary with the server version and account permissions.
| Tool | What it does |
|---|---|
| about_md_business | Explains what can be done in the open folder and which tools to call, in order, for each task. |
| list_schemas | Lists the document schemas available. |
| get_schema | Returns a document schema. |
| read_document | Reads a business document. |
| validate_document | Validates a document against its declared schema. |
| create_document | Creates a new document, validated against its schema. |
| update_document | Updates an existing document, validated against its schema. |
| search_documents | Searches Markdown documents and TSV test sheets in the open folder. |
| read_tsv | Reads a tab-separated test sheet. |
| append_tsv_row | Appends a row to a test sheet. |
| update_tsv_row | Updates one row of a test sheet, leaving other rows out of the diff. |
| check_tsv | Checks a test sheet without reading it row by row. |
| git_status | Shows git status for the open folder. |
| git_diff | Shows the diff of the agent's changes. |
| git_commit | Commits changes. There is no push tool. |
| read_data | Reads external JSON or XML as a tree, read-only. |
| data_to_table | Quotes a repeating part of JSON or XML data as a table. |
| read_har | Reads a HAR network capture. |
| search_lines | Finds matching lines in a log file; results are masked for tokens, cookies and email addresses. |
| read_lines | Reads the lines around a hit in a log file. |
| filter_records | Filters JSONL or TSV records. |
| aggregate | Aggregates JSONL or TSV records. |
| build_timeline | Merges several files into one time-ordered timeline. |
| save_evidence | Saves an extract as a file an investigation report can cite. |
| declare_web_mode | Declares the folder a website for site-file editing. |
| web_mode_status | Reports website mode status. |
| list_site_files | Lists the website's files. |
| read_site_file | Reads a website file. |
| write_site_file | Writes a website file, shown live in the app. |
| open_in_app | Opens a document in the desktop app. |
| list_open_documents | Lists the documents open in the app. |
| close_document | Closes an open document tab. |
| export_pdf | Opens a document and brings up the print dialog for A4 PDF. |
| capture_window | Captures an image of the app's own window. |
{
"mcpServers": {
"md-business": {
"type": "http",
"url": "http://127.0.0.1:<PORT>/md-business",
"headers": {
"Authorization": "Bearer <TOKEN>"
}
}
}
}In the md-business desktop app, open the folder and use MCP tab > Add settings to the open folder; Claude Code reads the resulting .mcp.json automatically. <PORT> and <TOKEN> are filled in by the app.
Read, search and edit files inside allowed directories with the Model Context Protocol reference Filesystem server.
Retrieve version-aware library documentation and code examples with Context7 MCP, using resolve-library-id and query-docs.
A powerful **Model Context Protocol (MCP) server** that exposes [Apktool](https://ibotpeaches.github.io/Apktool/) functionality for Android APK
This project provides an MCP (Model Context Protocol) server for PipeCD, enabling integration and management of applications and deployments.
A Swift framework for building AI agents with Amazon Bedrock and Model Context Protocol (MCP) support. AgentKit simplifies creating conversational AI
AI explainability and transparency reports for AI agents. Instant transparency scan, model c...