Deterministic data sanitization for AI agents via Model Context Protocol.
Bridging the gap between LLM intent and dataset integrity through a standardized terminal interface.
> Initializing connection to /mcp endpoint...
> Connection established. 10 tools registered.
01 // The Protocol
What is MCP?
The Model Context Protocol defines a host-client-server architecture. The host (Claude Desktop) initiates a JSON-RPC lifecycle, discovering server tools and granting the client (the LLM) context-aware capabilities without local code execution risks.
READ_OFFICIAL_DOCS02 // The Logic
Deterministic Only
We operate as a deterministic tools-only server. When an agent requests a cleaning action, we execute defined regex and algorithmic sanitization. The model suggests the intent; we handle the execution with strict risk annotations.
03 // Registered Tools
clean_dataset
Initial structural scan.
scan_dataset
PII & health check.
get_cleaning_plan
Proposed actions.
apply_cleaning_plan
Bulk modifications.
apply_cleaning_action
Single target fix.
preview_dataset
Head/tail data view.
export_dataset
Polished CSV output.
create_upload_url
Secure ingress flow.
delete_dataset
Permanent erasure.
list_cleaning_actions
Audit trail logs.
04 // Safety Constants
BEARER_AUTH
Standardized token verification for session entry.
ISOLATION
Logical separation of workflow runtimes.
TTL_EXPIRY
All artifacts purged automatically after 60 mins.
SSRF_GUARD
Restricted URL fetching prevents internal leaks.
ZERO_LOGS
Metadata logs only. Content never persisted.
05 // Quota Definition
| PARAMETER | FREE_TIER | PRO_TIER |
|---|---|---|
| DAILY_LIMIT | 5 FILES / DAY | UNLIMITED |
| MAX_PAYLOAD | 2MB (CSV) | 25MB (ALL FORMATS) |
| AUDIT_REPORTS | ΓÇö | ENABLED |
| REPLAY_SCRIPTS | ΓÇö | INCLUDED |