LeXML-mcp
MCP server for LexML Brasil — search the legislative SRU acervo, resolve URNs, and inspect service metadata.
README
LeXML-mcp
MCP server for LexML Brasil — search the legislative SRU acervo, resolve URNs, and inspect service metadata. Built with FastMCP + httpx.
Features
| Tool | Description |
|---|---|
lexml_search |
Search the LexML SRU acervo with CQL queries |
lexml_resolve_urn |
Resolve a LexML URN to its public URL |
lexml_explain |
Retrieve metadata about the SRU service |
Every response includes a provenance envelope (source_url, source_kind, authority, retrieved_at, content_hash, backend_name, backend_version, schema_version).
Quick start
# Install
uv sync
# Run the server (stdio transport)
uv run lexml-mcp
The server listens on stdio — configure your MCP client to launch it:
{
"mcpServers": {
"lexml": {
"command": "uv",
"args": ["run", "--directory", "/path/to/LeXML-mcp", "lexml-mcp"]
}
}
}
Usage examples
lexml_search
# Search for documents about consumer protection law
result = await lexml_search(
query='dc.title any "código de defesa do consumidor"',
start_record=1,
maximum_records=10,
record_schema="dc"
)
Returns structured records with Dublin Core fields, number_of_records, next_record_position, and provenance.
lexml_resolve_urn
# Resolve a URN to its public URL
result = await lexml_resolve_urn(
urn="urn:lex:br:federal:lei:1990-09-11;8078"
)
Returns public_url, status_code, redirect_chain, and provenance.
lexml_explain
# Inspect SRU service capabilities
result = await lexml_explain()
Returns SRU explain record data with server info, database info, and supported schemas.
Challenge detection
The SRU endpoint may return an HTML verification page instead of XML. The server detects this and returns a structured object:
{
"is_challenge_like": true,
"content_type": "text/html",
"raw_excerpt": "<!DOCTYPE html>...",
"status_code": 200,
"provenance": { ... }
}
Error handling
All errors are returned as structured objects — never crashes:
| Error type | Trigger |
|---|---|
network_error |
Connection refused, DNS failure |
timeout |
Request exceeded timeout (default 15s) |
invalid_xml |
SRU returned malformed XML |
http_error |
Non-2xx HTTP status |
Caching
LRU cache with configurable TTL (default 300s, max 256 entries). Challenges are never cached as successes. Configure via environment:
LEXML_CACHE_TTL— seconds (default300)LEXML_CACHE_MAXSIZE— max entries (default256)
Configuration
| Env variable | Default | Description |
|---|---|---|
LEXML_SRU_URL |
https://www.lexml.gov.br/busca/SRU |
SRU endpoint |
LEXML_URN_TEMPLATE |
https://www.lexml.gov.br/urn/{urn} |
URN resolution template |
LEXML_TIMEOUT |
15.0 |
HTTP request timeout (seconds) |
LEXML_CACHE_TTL |
300 |
Cache TTL (seconds) |
LEXML_CACHE_MAXSIZE |
256 |
Max cache entries |
Project structure
src/lexml_mcp/
├── server.py # FastMCP server, tool registration
├── config.py # Environment-based configuration
├── connectors/
│ └── acervo.py # SRU search, explain, URN resolution
├── models/
│ ├── provenance.py # Provenance envelope
│ ├── search.py # Search result models
│ └── error.py # Structured error objects
└── utils/
├── cache.py # LRU TTL cache
└── challenge.py # Challenge HTML detection
Testing
uv run pytest
29 tests covering:
- SRU success parsing (Dublin Core records)
- Challenge HTML detection
- Malformed XML handling
- Timeout simulation
- URN resolution
- Explain response parsing
- Provenance envelope construction
- Cache set/get/expiry/eviction
- Structured error types
- Tool registration
Limitations
- Parser, linker, renderer not implemented — these require JVM/Scala/Haskell infrastructure and are roadmap items
- SRU endpoint may return challenge HTML — the server detects and reports this transparently
- No multi-tenancy — single-user, local-first design
- No DOCX output — renderer integration is future scope
- No Unicode normalization — preserves original content as-is
- No external health checks — liveness is local only
Dependencies
mcp— official Python SDK (FastMCP)httpx— async HTTP client- stdlib only beyond those two
License
GPL-3.0-or-later
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.