Context
This is private client work, so code and sensitive organization details are not public. The work combined Redmine MCP tooling with a corporate VK Teams RAG bot over Wiki.js and PDF regulations.
Problem
Internal teams need agents that can work with existing systems without exposing messy raw APIs. They also need knowledge search that respects roles and source boundaries.
Constraints
- Private/client environment.
- Redmine actions had to be represented as safe, compact tools for an LLM agent.
- Search had to work across Wiki.js pages and PDF regulations.
- Answers had to account for access roles.
Approach
The Redmine side exposed a controlled MCP tool surface. The RAG side combined indexed fragments, hybrid retrieval, and VK Teams delivery for corporate use.
Implementation highlights
- 12 MCP tools for task, project, user, metadata, and reporting workflows.
- Compact JSON output, safe errors, pagination, sorting, and overdue task handling.
- Role-aware retrieval over Wiki.js and PDF sources.
- Hybrid search over 200+ documents and 2-3k indexed fragments.
Result
The work demonstrates practical enterprise AI integration: controlled tool surfaces, internal system access, and RAG over real corporate documents.
What it demonstrates
MCP design, enterprise RAG, private integration work, safe agent outputs, and internal assistant architecture.