Pasal.id — Indonesian Law MCP Server
The first open, AI-native Indonesian legal database. Gives Claude and any MCP-compatible AI assistant grounded, citation-ready access to 100,000+ Indonesian regulations — UU, PP, Perpres, PERMEN, PERDA, and 20+ other regulation types — directly from authoritative sources (peraturan.go.id, BPK JDIH, JDIH MK, JDIHN).
- Live server:
https://mcp.pasal.id/mcp - Transport: Streamable HTTP
- Auth: OAuth 2.0 — free account at pasal.id
- Setup in 30 seconds: pasal.id/hubungkan
- License: AGPL-3.0
Why this exists
Indonesian law is scattered across dozens of government portals, each with its own PDF quirks and status ambiguities. Most LLM answers about Indonesian regulations are hallucinated — the training data is thin, and the live web isn't indexed for precise legal retrieval.
Pasal.id consolidates 100,000+ regulations into structured data (pasal-level granularity, whole-law hierarchy, preamble sections, amendment history, and legal-status provenance) and exposes it to AI assistants via MCP. Every answer an LLM gives using this server can be grounded in a specific article, regulation part, and authoritative source.
Connect
Pick your client. All use the same server URL: https://mcp.pasal.id/mcp.
Claude Code — one command
claude mcp add --transport http pasal-id https://mcp.pasal.id/mcp
Claude Code will open a browser window for OAuth on first use. For CI/automation, use a personal access token from pasal.id/akun:
claude mcp add --transport http pasal-id https://mcp.pasal.id/mcp \
--header "Authorization: Bearer ${PASAL_MCP_TOKEN}"
Claude Desktop / Cursor / Windsurf / VS Code
All consume the same JSON. Paste into the relevant config file:
{
"mcpServers": {
"pasal-id": {
"type": "http",
"url": "https://mcp.pasal.id/mcp",
"headers": {
"Authorization": "Bearer ${PASAL_MCP_TOKEN}"
}
}
}
}
| Client | Config file |
|---|---|
| Claude Desktop | ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) · %APPDATA%\Claude\claude_desktop_config.json (Windows) |
| Cursor | ~/.cursor/mcp.json or .cursor/mcp.json (workspace) |
| Windsurf | ~/.codeium/windsurf/mcp_config.json |
| VS Code MCP | .vscode/mcp.json (workspace) |
| Claude Code | .mcp.json (project root) |
Full per-client walkthrough: docs/clients.md.
ChatGPT Desktop (April 2026+)
Settings → Developer Tools → Add MCP Server → paste https://mcp.pasal.id/mcp.
Tools
The server exposes six task-oriented tools to AI assistants (v2.0.0 — the earlier eleven-tool surface keeps working for existing integrations but is deprecated and scheduled for removal after August 2026):
| Tool | Purpose |
|---|---|
search_legal | Find regulations when you do not yet know which law applies — topics, keywords, or messy citations, with validated granular filters (multiple types, year, status, issuing body, region) and diagnostics that explain zero results. |
resolve_law | Turn any reference — "KUHP", "UU 13/2003", "Pergub DKI 220 tahun 2010", a pasted title — into a canonical law_id, or get disambiguation candidates instead of a guess. |
get_law_context | Compact orientation on one law: summary (status, structure counts, top-level outline), a pasal-range-compressed outline, or typed relationship groups (amendments, court reviews) — token-budgeted. |
read_law | Read text with forgiving selectors: pasal 27, pasal 27-30, bab III, menimbang, penjelasan pasal 5, lampiran, all — with cursors, char budgets, and explicit missing reporting. |
search_court_decisions | Search Constitutional Court (Mahkamah Konstitusi) decisions by reviewed law, lane (PUU/SKLN/PHPU/PHPKADA), ruling, year, dissent, or judge. |
report_issue | Report data problems; the search_failure type ("I searched X and expected law Y") feeds the search regression suite directly. |
Every v2 response carries a request_id and structured diagnostics; errors are
typed (error_code + recovery guidance + valid_values) so assistants can
self-correct instead of retrying blindly. The canonical workflow: resolve_law
when you know the law → get_law_context to orient → read_law for text;
search_legal when you don't.
Schemas + example calls: docs/tools.md.
Example prompts
Once connected, ask your AI assistant questions like:
- "Explain Pasal 81 of the UU Cipta Kerja on employment."
- "Is the 1974 Marriage Law still in force?"
- "What are contract workers' rights under Indonesian law?"
- "Compare worker rights before and after the Job Creation Law."
- "What is the minimum marriage age in Indonesia?"
- "Tampilkan struktur UU ITE dan ambil bagian Menimbang serta Pasal 27."
- "Cari dalam UU PDP bagian tentang pengendali data, lalu cek status peraturannya."
- "Saya menemukan OCR yang salah di Pasal ini; laporkan koreksinya ke Pasal.id."
The assistant should search first, retrieve only the needed Pasal or regulation part, verify status/provenance when the answer depends on currency, and use report_issue when the parsed corpus has a real data problem.
About the data
- Primary source: peraturan.go.id (Ministry of State Secretariat).
- Secondary sources: BPK JDIH (69 regulation types), JDIH MK (Constitutional Court), JDIHN (regional regulations).
- Verification: Gemini 3 Flash cross-checks every parsed document against the source PDF. Unverified regulations are flagged.
- Updates: New regulations ingested within 24–48 hours of publication.
- Gaps: See pasal.id/metodologi for known limitations.
Contributing
Pasal.id itself lives in a private repository — but this repo is public and we welcome:
- Corrections to the public MCP surface (README, install snippets, docs).
- Bug reports against the live server via GitHub Issues or pasal.id/masukan.
- Feature requests — especially new MCP tools or improvements to existing ones.
PRs to the source regulations database happen through the crowd-correction workflow at pasal.id: any user can highlight a passage and suggest a fix; our team reviews and merges within days.
Related projects
- pasal.id — web reader with full search, law detail pages, and crowd corrections.
- pasal.id/api — REST API (same data, same authentication, for non-MCP integrations).
License
AGPL-3.0 — see LICENSE.
The MCP server code and the regulation data are both open. Attribution is appreciated; commercial derivatives must remain open-source under the same license.
Built by Aturio · Maintained by Ilham Firdausi Putra.