Real-time legal intelligence substrate for Brazilian lawyers. Monitors large-scale Brazilian court case data (CNJ integrations), classifies legal risk using a 5-agent pipeline in real time, and exposes structured intelligence via MCP so any AI agent can act on it.
All components (agents, MCP tools, audit logs, and traces) originate from a single execution layer. ProcessRadar was designed for one of the most complex judicial environments in the world.
Unlike jurisdictions that provide centralized court APIs, Brazil operates through hundreds of independent judicial systems, each with different architectures, standards, and publication formats.
ProcessRadar was engineered specifically to operate under these constraints.
The platform continuously monitors and normalizes information from multiple heterogeneous sources, including:
- PJe
- e-SAJ
- Projudi
- Official Court Gazettes
- State and Federal Court Portals
This requires transforming highly inconsistent procedural data into a unified structure that can be analyzed reliably by AI agents.
Legal data often contains highly sensitive personal information.
To minimize exposure and comply with LGPD principles, ProcessRadar follows a data-minimization architecture:
- Sensitive source documents are processed and discarded whenever possible.
- Only the minimum operational metadata required for monitoring is retained.
- Law-firm data is isolated through Row-Level Security (RLS).
- MCP tools enforce additional access controls and guardrails.
- Segredo de Justiça cases trigger dedicated protection flows.
Legal monitoring generates large volumes of procedural events, many of which do not require deep analysis.
ProcessRadar uses a multi-stage routing architecture that:
- Filters routine procedural updates through lightweight pipelines.
- Escalates only relevant events to advanced Gemini models.
- Generates structured legal intelligence, risk signals, and actionable summaries.
This approach significantly reduces computational cost while maintaining analytical quality.
Rather than storing every uploaded legal document indefinitely, ProcessRadar follows a privacy-by-design approach:
- CNIS, PPP and similar source files are processed and discarded after extraction.
- Generated petitions and agreements remain under the lawyer's control.
- Only auditable metadata and operational records are retained when necessary.
The objective is simple:
the most secure sensitive document is the one that was never permanently stored.
Brazilian legal professionals spend countless hours monitoring court systems, reviewing procedural movements, identifying deadlines, and manually extracting relevant information from judicial updates.
ProcessRadar automates this workflow.
The platform continuously monitors court activity and transforms raw procedural movements into structured intelligence such as:
- Case summaries
- Risk detection
- Deadline identification
- Priority alerts
- Procedural insights
- Legal guidance support
Each update flows through a deterministic pipeline enhanced by AI analysis and multiple validation layers.
- Supabase (PostgreSQL)
- Supabase Edge Functions
- Row-Level Security (RLS)
- Audit Logging
- MCP Server
- MCP Gateway (Authentication, Rate Limiting, Audit)
- Gemini 3.5 Flash
- Gemini 2.5 Pro
- Vertex AI
- React
- TypeScript
- Vite
- Tailwind CSS
- Multi-tenant architecture
- Privacy by Design
- Data Minimization
- Defense in Depth
- AI-assisted, human-centered workflows
ProcessRadar operates through specialized agents:
Synchronizes judicial data from external providers and court sources.
Continuously monitors procedural movements and identifies relevant changes.
Transforms procedural updates into structured summaries and legal insights.
Identifies deadlines, procedural risks, missing documentation, and critical events.
Generates alerts and delivers actionable information to legal professionals.
ProcessRadar exposes structured legal intelligence through the Model Context Protocol (MCP).
Available tools:
Public mode (demo token, no login required):
get_platform_overviewget_agent_architectureget_guardrailsget_demo_metrics
Authenticated mode (per-user bearer token, scoped by RLS):
list_processesget_process_summary
Public mode serves frozen snapshot data; authenticated mode hits the live DB under RLS. The two restricted tools above are intentionally blocked for the public demo token and will return an authorization error if called without a per-user token.
This allows external AI systems and agent frameworks to securely consume ProcessRadar intelligence through standardized MCP interfaces.
Potential consumers include:
- Claude Desktop
- Google ADK Agents
- OpenAI Agents
- Enterprise automation platforms
- Custom AI workflows
- Try MCP Live → https://processradar.com.br/mcp/playground
- Connect Google ADK →
examples/google-adk-agent.py - Architecture →
AGENTS.md
MCP playground connected to the production MCP gateway. In public mode the gateway serves frozen snapshot data; in authenticated mode it queries the live database under RLS.
ProcessRadar follows a strict data minimization philosophy.
Unlike traditional legal platforms that store large volumes of sensitive client documents indefinitely, ProcessRadar processes information, returns results, and discards unnecessary data whenever possible.
Only data required for platform operation:
- User account information
- Subscription records
- Monitored legal processes
- Procedural movements
- Alerts and agenda items
- Audit logs
- Electronically signed documents (for legal validity)
Whenever technically possible, ProcessRadar avoids permanent storage of sensitive legal content:
- Generated petitions after download
- Generated contracts after export
- Temporary negotiation content
- Unsaved simulations
- Raw CNIS files
- Raw PPP files
- Raw LTCAT files
Uploaded documents used for extraction are converted into structured data and discarded after processing according to retention policies.
The platform follows the principles established by Brazil's General Data Protection Law (LGPD), especially:
- Data Minimization
- Purpose Limitation
- Necessity
- Accountability
- Row-Level Security (RLS)
- Workspace isolation
- Audit logging
- Segredo de Justiça filtering
- IP-based validation
- MCP authentication gateway
Legal documents are treated as untrusted content, not as executable instructions.
ProcessRadar applies defensive controls before any AI analysis, including input sanitization, structured tool-calling, schema-constrained outputs and instruction-boundary enforcement. Commands embedded inside user messages, uploaded PDFs, scanned documents or extracted text are never treated as system instructions.
This design choice became increasingly important as prompt-injection attacks began appearing in real-world legal and enterprise environments. The platform assumes that any document may contain adversarial content and isolates document text from operational instructions provided to the AI agents.
As a result, procedural documents can influence the legal analysis, but they cannot modify agent behavior, override platform rules, trigger unauthorized actions or change execution policies.
ProcessRadar includes domain-specific safeguards designed for legal environments.
Processes protected by judicial secrecy are automatically restricted from AI analysis.
Deadline calculations respect:
- Brazilian legal calendar
- Business days
- Court schedules
- Jurisdiction-specific rules
Actions are prevented when procedural conditions indicate execution would be unsafe or legally inappropriate.
Only the minimum required information is processed and retained.
ProcessRadar is available as a Progressive Web App (PWA), enabling installation on desktop and mobile devices with a near-native experience.
The platform is live in production and ready for evaluation.
To preserve the integrity of audit logs and the security of MCP agents, automated 1-click test access is provided exclusively through the official Google Hackathon submission panel. This ensures that only registered judges and competitors can reach the demo account, while keeping public repositories free of credentials that could be harvested by bots.
The demo account is fully populated with simulated legal processes and procedural data — no real client information is included.
Submitted under the Optimize Existing Agents track.
- Multi-agent orchestration
- MCP integration
- AI-assisted legal workflows
- Privacy by Design
- Production-oriented reliability
- Legal domain automation
ProcessRadar was designed to reduce the operational burden of legal monitoring.
The platform helps legal professionals:
• Reduce time spent reviewing procedural updates • Detect deadlines earlier • Centralize fragmented court information • Minimize manual monitoring workflows • Increase productivity through AI-assisted analysis
By transforming procedural data into actionable intelligence, ProcessRadar allows legal teams to focus on legal strategy rather than repetitive monitoring tasks.
ProcessRadar combines deterministic legal workflows, multi-agent orchestration, Gemini-powered analysis, and privacy-first architecture to help legal professionals monitor judicial proceedings more efficiently.
The platform was designed not only for human users, but also for the emerging ecosystem of AI agents through standardized MCP integrations, enabling secure access to structured legal intelligence while respecting privacy, compliance, and legal constraints.