Skip to main content

pk-eli-mcp

MCP server for Pakistani law: 967 federal statutes and 1,414 Supreme Court of Pakistan judgments, full text, with a strict citation contract on every response.

Pakistan publishes no machine-readable legal API, so this connector is corpus-based: it serves two public HuggingFace mirrors of official sources instead of querying a government endpoint live.

Collection Source Snapshot License
Federal statutes (967) Pakistan_Laws_Dataset, collected from pakistancode.gov.pk PDFs (Ministry of Law and Justice) 2025-01-30 ODC-BY 1.0
Supreme Court judgments (1,414) supreme-court-of-pak-judgments 2024-07-26 MIT

Honest limits, stated up front: both collections are static snapshots. Amendments after 2025-01-30 and judgments after 2024-07-26 are absent, and the judgment corpus is a subset of the Court's output. Every response repeats this in dataset_note. Verify currency against pakistancode.gov.pk and scp.gov.pk before relying on a provision.

Tools

Tool What it does
pk_search_laws Keyword search over the 967 federal statutes (title-weighted). First call downloads the 47 MB corpus once, then serves from cache.
pk_get_law Full statute text by law_id (truncated at ~300,000 characters).
pk_case_search Server-side full-text search over the Supreme Court judgments.
pk_get_decision One judgment's full text, by row_idx or registry case_id.
pk_verify_citations Anti-hallucination check: extracts Pakistani legal citations from any text and verifies each against the corpus.
pk_coverage Declare what this connector covers, when each family was captured, and - explicitly - what it does NOT cover. Every gap carries a fallback.

Citation verification (pk_verify_citations)

Run it on a drafted answer, memo or pleading before it reaches a reader. The tool extracts every checkable citation and verifies it against the corpus:

  • statute sections: section 302 of the Pakistan Penal Code, section 10A of the Pakistan Study Centres Act, 1976, sections 6 and 7 of ..., sub-section (1) of section 6 of ...;
  • act coordinates: Act No. XLV of 1860, Ordinance No. LII of 2000 (Roman or Arabic);
  • Constitution articles: Article 184(3) of the Constitution (the tail is required - a bare Article 5 could be any instrument and is deliberately ignored);
  • Supreme Court registry citations: Crl.A. 93/2013, Criminal Appeal No. 93 of 2013.

The semantics are deliberately hard:

  • A cited section that does not exist makes the whole result HALLUCINATION_DETECTED with isError=true, plus a range hint of what does exist (section 999 does not exist in this act; 59 sections are machine-detectable in its text, numbered section 1-66).
  • NO_CITATIONS_FOUND is not a success - it means there was nothing to check.
  • A parenthetical description right after a citation (section 302 of the Pakistan Penal Code (punishment of qatl-i-amd)) is content-checked against the real provision text with a character-trigram match. A mismatch is a review signal, never a hard block.
  • Everything the tool could NOT check is listed in a structured gaps field. The judgment corpus is a subset and roughly half the statute PDFs carry no machine-readable coordinate header, so an unresolvable citation is reported as out_of_corpus ("existence UNKNOWN, not disproven"), never as a hallucination.

The parse-verify-report pattern is adapted from korean-law-mcp (MIT) - see THIRD_PARTY.md. The same pattern shipped first in it-eli-mcp v0.6.0; the fleet rollout plan lives in docs/VERIFY-ROLLOUT.md.

Citation contract

Pakistan has not deployed ELI. Every response still carries the fleet contract:

  • eli_uri - a stable, resolvable URL: the canonical pakistancode.gov.pk PDF for statutes, the dataset row URI for judgments. Never invented; see eli_note.
  • human_readable_citation - e.g. THE PRIVATISATION COMMISSION ORDINANCE, 2000 or Supreme Court of Pakistan, Crl.A. 93/2013.
  • source_url - where a human can read the same document.

Install

uvx pk-eli-mcp          # or: pip install pk-eli-mcp

Claude Code:

claude mcp add pk-eli-mcp -- uvx pk-eli-mcp

Or drop .mcp.json.example into your project as .mcp.json.

Windows 11 with Smart App Control

Smart App Control blocks unsigned executables, which covers uvx.exe, pip.exe and the pk-eli-mcp.exe launcher that pip writes at install time. The python.exe and py.exe from the python.org installer are signed by the Python Software Foundation, so running the module through the interpreter works:

python -m pip install pk-eli-mcp
python -m pk_eli_mcp

pip.exe is blocked for the same reason, so install with python -m pip, not pip install. If python is not on PATH, use the Windows launcher: py -3 -m pk_eli_mcp.

{ "mcpServers": { "pk-eli-mcp": { "command": "python", "args": ["-m", "pk_eli_mcp"] } } }

Do not turn Smart App Control off to work around this - it cannot be re-enabled without reinstalling Windows.

Configuration

Env Default Purpose
PK_ELI_CACHE_DIR ~/.matematic/cache/pk-eli Corpus file + HTTP cache
PK_ELI_AUDIT_DIR ~/.matematic/audit JSONL audit log (one line per tool call)

Development

python -m venv .venv && .venv/Scripts/pip install -e ".[dev]"   # Linux/macOS: .venv/bin/pip
ruff check src tests
pytest tests/test_instructions_drift.py tests/test_corpus.py   # offline
pytest tests/test_smoke.py                                     # live (downloads the corpus)

Governance

Read-only by design: the server can search and fetch, never write. Every tool call appends an audit line (~/.matematic/audit/pk-eli-mcp.jsonl) with a SHA-256 input hash, duration and status. See CONSTITUTION.md for the rules this connector is built under and DISCOVERY.md for how the sources were chosen.

License

Apache-2.0 for this code. The datasets keep their own licenses (ODC-BY 1.0 and MIT); the underlying texts are Government of Pakistan public legal documents.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pk_eli_mcp-0.4.0.tar.gz (159.1 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

pk_eli_mcp-0.4.0-py3-none-any.whl (44.0 kB view details)

Uploaded Python 3

File details

Details for the file pk_eli_mcp-0.4.0.tar.gz.

File metadata

  • Download URL: pk_eli_mcp-0.4.0.tar.gz
  • Upload date:
  • Size: 159.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for pk_eli_mcp-0.4.0.tar.gz
Algorithm Hash digest
SHA256 89c56cff5c279a0ffaf3996a68fcc7e27d6b7632e232a6d3dbf5e256fd33e899
MD5 f702c8f84b7956e57e378bc55ed03fb1
BLAKE2b-256 ad7a9f54d2d4bdfab04699b54c119aa7dcb1b09082a21e7ad5f2be064d2bfbfc

See more details on using hashes here.

File details

Details for the file pk_eli_mcp-0.4.0-py3-none-any.whl.

File metadata

  • Download URL: pk_eli_mcp-0.4.0-py3-none-any.whl
  • Upload date:
  • Size: 44.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for pk_eli_mcp-0.4.0-py3-none-any.whl
Algorithm Hash digest
SHA256 c5d4bfa87a1e9b025f56e671311b9a4e262bb24e2bc498457522dc37bd2f99ee
MD5 87a2a9b84a59fb5c6a973ef01d9659fb
BLAKE2b-256 085272ef49415b567e0cfa4638277a269d80e9d30810eef0f4f64738b98d2d89

See more details on using hashes here.

Release history Release notifications | RSS feed

0.4.3

2 files

0.4.2

2 files

0.4.1

2 files

This release

0.4.0 This release

2 files

0.3.4

2 files

0.3.3

2 files

0.3.2

2 files

0.3.1

2 files

0.3.0

2 files

0.2.0

2 files

0.1.0

2 files

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page