MCP server for Reporium — query your AI repo library from Claude
reporium-mcp
reporium-mcp is the MCP server for the Reporium platform. It turns the live portfolio into a tool surface that agents can query directly from Claude Code and other MCP-aware clients.
Why MCP?
MCP makes the portfolio AI-native instead of screen-native.
Without MCP, an agent has to infer state from documentation or ask a human to look things up. With MCP, the agent can query the live portfolio directly:
- search for repos by keyword or semantics
- inspect a repo in detail
- browse taxonomy dimensions and values
- answer portfolio questions in natural language
- surface portfolio gaps and trend signals as part of an execution loop
That means portfolio intelligence becomes a first-class tool in agent workflows instead of a manual reference step.
Installation
pip install -r requirements.txt
Configuration
Copy .env.example to .env and fill in your values:
cp .env.example .env
REPORIUM_API_URL=https://your-reporium-api-url
REPORIUM_API_KEY=your-api-key-here
REPORIUM_API_KEY is optional if the target API is publicly reachable for the endpoints you use.
Adding To Claude Code
claude mcp add reporium -- python /path/to/reporium-mcp/mcp_server.py
Replace /path/to/reporium-mcp/ with the absolute path where you cloned this repo.
Tool Surface
This server exposes 13 tools.
| Tool | What it does | Core parameters |
|------|--------------|-----------------|
| search_repos | Keyword search over repo names, descriptions, and other indexed text fields | query, limit |
| search_repos_semantic | Embedding-based semantic repo search with similarity scores | query, limit |
| get_repo | Full repo detail including taxonomy, skills, categories, commits, and metadata | name |
| find_similar_repos | Finds repos similar to a target repo by reusing its summary as a semantic query | repo_name, limit |
| list_taxonomy_dimensions | Lists active taxonomy dimensions with repo coverage counts | none |
| list_taxonomy_values | Lists available values for a specific taxonomy dimension | dimension |
| get_repos_by_taxonomy | Returns repos matching a taxonomy dimension/value pair | dimension, value, limit |
| ask_portfolio | Sends a natural-language portfolio question to the Reporium intelligence layer | question |
| get_portfolio_gaps | Returns portfolio gap analysis for under-covered areas | none |
| get_ai_trends | Returns trend-oriented signals from the current portfolio snapshot | none |
| get_portfolio_insights | Returns proactive portfolio signals including rising gaps, stale repos, velocity leaders, and near-duplicate clusters | none |
| get_cross_dimension_stats | Returns counts at the intersection of two taxonomy dimensions | dim1, dim2, limit |
| get_repo_quality | Returns only a repo's quality signals payload for lightweight quality checks | name |
Supported taxonomy dimensions include:
skill_areaindustryuse_casemodalityai_trenddeployment_context
Example Questions
Once the MCP server is installed in Claude Code, you can ask:
- "What repos do we have for healthcare NLP?"
- "Find me production-ready RAG frameworks in the library"
- "Show me repos similar to langchain"
- "List all taxonomy dimensions and their most common values"
- "What are the current portfolio gaps?"
- "What AI trends are rising in the portfolio?"
- "Which repos cover deployment context for on-device inference?"
Relationship To The Rest Of Reporium
reporium-apiis the live data and intelligence backendreporiumis the human-facing portfolio UIreporium-mcpis the agent-facing tool layer
All three should describe the same portfolio, just through different interfaces.