dataesr-mcp

Model Context Protocol (MCP) server for interacting with dataesr resources.

Features

• scanR: Search across multiple scanR indexes (publications, persons, organizations, projects, participations, patents)
• Affiliation Matcher: Match affiliation strings against reference systems (ROR, GRID, RNSR, etc.)
• Flash RAG: Retrieve relevant documents from French Ministry of Higher Education and Research publications
• Tableaux prototyping: Serve the dashboard prototyping guidance of dataesr/tableaux (AGENTS.md files, README, skills), its board templates and data-source index, and validate a dashboard manifest against the repository's JSON Schema, without executing generated code.

The tableaux tools are registered at /tableaux and will be available after the service is deployed (using the same authentication as the other mounted MCP servers): list_guidance, get_guidance, list_templates, get_template, list_data_sources, get_data_source_guide, list_dashboard_components, recommend_architecture, and validate_dashboard_spec.

Nothing about the tableaux conventions is stored in this repository. Each call reads the current files from GitHub through the contents API (cached for 5 minutes), starting from the index docs/dashboard-prototype/prototype.index.json, which lists the guidance documents, template paths, components and data sources. Change the content in dataesr/tableaux, not here. Configuration (environment variables):

Variable	Default	Purpose
TABLEAUX_REPO	dataesr/tableaux	Repository to read
TABLEAUX_REF	staging	Branch, tag, or commit to read
TABLEAUX_PROTOTYPE_INDEX_PATH	docs/dashboard-prototype/prototype.index.json	Location of the index in the repository
GITHUB_TOKEN	unset	Optional read-only token; raises the GitHub rate limit (60 requests/hour unauthenticated) and is required for a private repository

validate_dashboard_spec checks the manifest against the repository's schema plus the rules JSON Schema cannot express (route paths belong to the board, data source ids exist in the index). It does not validate, build, or run a submitted project. Do not execute generated code in the MCP service; isolated project validation requires a separately deployed sandbox.

Authentication

All MCP endpoints accept either the static bearer token (MCP_AUTH_TOKEN) or an OAuth 2.1 access token (authorization code + PKCE, refresh tokens, client_credentials). Tokens are HMAC-signed with MCP_AUTH_TOKEN (no storage; changing it revokes everything).

Two ways to connect an OAuth client (e.g. Claude):

• Pre-configured client (recommended): in the connector's advanced settings, set OAuth Client ID to MCP_OAUTH_CLIENT_ID (default dataesr-mcp) and OAuth Client Secret to the value of MCP_AUTH_TOKEN. The user then only clicks "Allow" on the consent page.
• Dynamic registration (no credentials configured): the client registers itself, and the user types MCP_AUTH_TOKEN on the consent page.

Set MCP_PUBLIC_URL (e.g. https://mcp.example.org) if the server is behind a proxy that doesn't forward Host/X-Forwarded-* headers.

Setup

Prerequisites

• Python 3.12+
• Docker (optional)

Local Development

# Install dependencies
uv sync

# Copy and configure environment variables
cp .env.example .env
# Edit .env with your credentials

# Run server
uv run main.py

Docker

# Build and run
make build
docker run -p 8000:8000 --env-file .env dataesr/dataesr-mcp:latest

# Or using docker-compose
docker-compose up

Usage

Start MCP Inspector:

npx -y @modelcontextprotocol/inspector

Deployment

The staging deployment is automated via GitHub Actions.

Or manually:

# Build and push
make build
make push