Purpose
diadoc-mcp-ru is an MCP (Model Context Protocol) server that exposes the Kontur Diadoc EDI API for use by AI assistants, enabling AI to handle common tasks in electronic document exchange: viewing sent and received documents, tracking document flow status, managing counterparties and their EDI connection invitations, performing signatures, handling machine-readable powers of attorney (МЧД), and obtaining printed forms.
Main capabilities
The interface catalog compiled from the official documentation developer.kontur.ru/doc/diadoc-api is stored in the repository in diadoc_mcp/endpoints.yaml, with a total of 114 methods, of which 78 are reads, 29 are writes, and 7 are irreversible operations. The server strictly follows this file, so the method list and the code remain consistent.
Typical use cases include:
- Displaying incoming documents that need to be processed;
- Summarizing document flow with a counterparty over a certain period;
- Finding a counterparty by ИНН and КПП, and checking whether it is already connected to EDI;
- Downloading printed forms of documents for finance use.
How to use
Two installation paths are provided, choose according to whether you use a terminal:
- No terminal: download the diadoc-mcp-ru-vX.Y.Z.mcpb file from the releases page and double-click to open it; Claude Desktop will automatically install the server and ask for the key in a separate window, with no need to manually edit configuration files. Python 3.10 or newer is required; on first launch the server installs dependencies itself.
- Manual: configure in Claude Desktop's claude_desktop_config.json; for the specific configuration contents, refer to the repository instructions.
An application identifier must be requested by sending an email to Kontur; production requires a license, while testing can use the test environment. A user token is obtained via the Authenticate method using a login name and password or a certificate. The key does not have to be placed in environment variables; the server supports a "cabinets" mechanism, storing it in ~/.ru-mcp/cabinets.json, with permissions 600, outside the repository.
The server provides three tools instead of 114 functions: the AI agent searches for methods in natural language, reads method cards, and invokes them. Read operations are executed directly; write and irreversible operations require confirmation. The authorization header does notleave the service domain, even when arbitrary paths are called.
Checking the installation: running the relevant command prints the number of loaded methods, whether the key was found and its source, and does not display secret contents; adding --live performs one low-cost real read call.
Privacy and license
The server runs on the user's machine, the key is stored locally in ~/.ru-mcp/cabinets.json, requests are sent only to the Diadoc (Kontur) API, there is no telemetry, and no user data is collected or transmitted. For the full privacy statement, see PRIVACY_POLICY.md.
The project uses the MIT license, and the author is Ilya Utov. The core has been extracted as schema-mcp-core; related servers include vk-mcp-ru, sbis-mcp-ru, chestny-znak-mcp-ru; e-commerce platform projects are marketplaces-mcp-ru. All projects can be viewed by purpose at ilyautov.github.io.
