Pyproc

PyProc MCP

Real-time Indonesian procurement data for LLM agents

---

PyProc MCP turns public SPSE/Inaproc procurement data into MCP tools that can be used by LLM clients (Claude Desktop, Continue, Cursor), AI agents, automation workflows, Python scripts, and command-line users.

Why PyProc MCP?

Indonesia's public procurement system (SPSE/Inaproc) processes thousands of tender and non-tender packages across hundreds of government institutions (LPSE). This data is public — but not easily accessible to AI agents and automation tools.

PyProc MCP bridges this gap:

• LLM-native access — MCP tools let AI agents search and analyze procurement data directly
• Real-time data — Fetches live data from SPSE/Inaproc, not stale snapshots
• Comprehensive — Full package details: announcements, participants, evaluation results, winners, schedules
• Lightweight — Single command: pyproc-mcp. Works with any MCP-compatible client
• Multi-interface — Use as MCP server, Python library, or CLI tool — same codebase, no lock-in
• Respectful defaults — Built-in rate limiting, caching guidance, and responsible-use policy

What You Can Do

• Discover opportunities — Search tender and non-tender packages across any LPSE host
• Deep-dive analysis — Retrieve full package details: requirements, HPS value, location, schedule
• Competitive intelligence — See who's bidding, who's winning, and at what price
• Timeline tracking — Monitor procurement schedules from announcement to contract
• Vendor research — Find which companies win which types of contracts where
• AI-powered insights — Let LLMs analyze procurement patterns, compare packages, and generate reports
• Export and automate — Use the Python library or CLI to build automated procurement monitoring pipelines

---

Usage Modes

PyProc can be used in three ways. Choose the one that fits your workflow:

1. 🧠 MCP Server for LLM Clients (Recommended)

Give your LLM direct access to Indonesian procurement data through MCP tools.

PyProc MCP supports two transport interfaces: stdio (default, for local MCP clients like Claude Desktop and Cursor) and HTTP (Streamable HTTP, for network/containerized deployments).

pip install pyproc[mcp]

# Start MCP server (stdio, default)
pyproc mcp

# Start MCP server (HTTP)
pyproc mcp --interface http
pyproc mcp --interface http --port 9090

Generate MCP client configuration:

The quickest way to connect your MCP client is to generate a ready-to-use config:

# Print config to stdout (stdio interface)
pyproc mcp --generate-config

# Print config to stdout (HTTP interface)
pyproc mcp --interface http --generate-config

# Write config to a file
pyproc mcp --generate-config mcp-config.json
pyproc mcp --interface http --port 9090 --generate-config mcp-config.json

Adding to your MCP client:

Copy the output into your client's config file (e.g. claude_desktop_config.json for Claude Desktop, or .cursor/mcp.json for Cursor). For stdio:

{
  "mcpServers": {
    "pyproc": {
      "command": "pyproc-mcp",
      "env": {
        "PYPROC_TIMEOUT": "30"
      }
    }
  }
}

For HTTP (clients that support Streamable HTTP):

{
  "mcpServers": {
    "pyproc": {
      "type": "streamableHttp",
      "url": "http://localhost:8080/"
    }
  }
}

Restart your MCP client. PyProc tools will appear in the tool list.

2. 📦 Python Library

Use PyProc as a Python library for custom automation and data analysis.

from pyproc import Lpse, JenisPengadaan
from pyproc.lpse import By

# Initialize client for a specific LPSE host
lpse = Lpse('kemenkeu')

# Search for tender packages
packages = lpse.get_paket_tender(
    start=0,
    length=20,
    search_keyword='mobil dinas',
    tahun=2025,
    kategori=JenisPengadaan.PENGADAAN_BARANG
)
print(packages['data'])

# Get full detail of a package
detail = lpse.detil_paket_tender('10080116000')
detail.get_all_detil()
print(detail.todict())

# Get only the winner
winners = detail.get_pemenang()
print(winners)

See [Python Library Usage](#python-library-usage) for complete API documentation.

3. ⌨️ CLI Tool

Download procurement data in bulk from the command line.

# Download tender data from Kemenkeu LPSE, export as JSON
pyproc spse kemenkeu --keyword "mobil dinas" --tahun-anggaran 2025 --output-format json

# Download non-tender data from multiple LPSE hosts
pyproc spse jakarta,sumbarprov --jenis-paket non_tender --tahun-anggaran 2025

# Download pencatatan non-tender data. This is distinct from non_tender.
pyproc spse nasional --jenis-paket pencatatan_non_tender --tahun-anggaran 2026

# Download with custom output filename
pyproc spse "kemenkeu;output_kemenkeu" --output-format csv --separator ";"

The spse subcommand is the default, so pyproc kemenkeu is shorthand for pyproc spse kemenkeu.

See [CLI Usage](#cli-usage) for the full argument reference.

---

MCP Tools

The MCP server exposes the following tools:

| Tool | Description | |---|---| | search_lpse_hosts | Find LPSE host slugs from institution names such as "kementerian keuangan" | | get_lpse_host_detail | Confirm metadata and canonical URL for a known LPSE host slug | | get_procurement_search_options | Explain direct keyword search vs local full-text indexing | | search_tender_packages | Search tender procurement packages by one or more exact SPSE keywords | | search_non_tender_packages | Search non-tender packages by one or more exact SPSE keywords | | search_pencatatan_non_tender_packages | Search pencatatan non-tender packages, a distinct /dt/nonspk entity | | search_swakelola_packages | Search swakelola packages | | search_pengadaan_darurat_packages | Search pengadaan darurat packages | | get_tender_detail | Get full detail for a tender package — announcement, participants, evaluations, winner, schedule | | get_non_tender_detail | Get full detail for a non-tender package | | get_pencatatan_non_tender_detail | Get detail for a pencatatan non-tender package | | get_swakelola_detail | Get detail for a swakelola package | | get_pengadaan_darurat_detail | Get detail for a pengadaan darurat package | | get_tender_details_bulk | Get details for multiple tender packages in one tool call | | get_non_tender_details_bulk | Get details for multiple non-tender packages in one tool call | | get_pencatatan_non_tender_details_bulk | Get details for multiple pencatatan non-tender packages | | get_swakelola_details_bulk | Get details for multiple swakelola packages | | get_pengadaan_darurat_details_bulk | Get details for multiple pengadaan darurat packages | | get_procurement_categories | List supported procurement categories (no network call) | | get_master_klpd | Get LKPP Satu Data master K/L/PD references (alternative data source) | | get_master_lpse | Get LKPP Satu Data master LPSE references (alternative data source) | | get_tender_umum_publik | Get tender data from LKPP ISB Satu Data (alternative data source / fallback) | | set_ssl_verify | Enable/disable TLS/SSL certificate verification for all SPSE requests | | get_ssl_verify | Return current SSL verification setting | | validate_lpse_host | Check if an LPSE host is accessible | | create_procurement_search_index | Download a bounded package set into a local SQLite full-text index | | search_procurement_index | Search a local SQLite full-text index | | list_procurement_indexes | List local full-text indexes | | delete_procurement_index | Delete a local full-text index |

Each tool includes LLM-friendly descriptions with parameter documentation, rate limit notes, and data source attribution.

---

LPSE Host Discovery

Users do not need to know LPSE host slugs in advance. When a user names an institution, the LLM should resolve the host first, then search procurement packages.

There are two procurement executor scopes:

• Agency-specific hosts, such as kemenkeu, jakarta, or pu.
• The nationwide host nasional, used for national-wide/lintas instansi data and pencatatan nasional sources.

If the user asks for national, nationwide, all-Indonesia, lintas instansi, or pencatatan nasional data, use lpse_host="nasional" directly.

Host discovery uses the maintained Gist host metadata and the newUrlPath field for agency hosts. The nasional host is built in and does not depend on the Gist metadata. Legacy oldUrl values are ignored. Canonical URLs are always built as:

https://spse.inaproc.id/{newUrlPath}

Example user request:

> cari data lelang pengadaan laptop pada kementerian keuangan

Expected MCP flow:

1. Call search_lpse_hosts with query="kementerian keuangan".
2. Select the best returned host, usually kemenkeu.
3. Call search_tender_packages with lpse_host="kemenkeu" and keyword="laptop".

If multiple host candidates look plausible, the LLM should ask the user to choose the intended LPSE before searching package data.

---

Search Modes

SPSE/Inaproc package search is keyword-based. PyProc MCP therefore exposes two search strategies:

• Direct keyword search: fast and lightweight. Use search_tender_packages or search_non_tender_packages with keyword or keywords, for example ["laptop", "notebook", "komputer"]. The MCP server runs bounded SPSE searches and merges duplicate package IDs.
• Local full-text search: broader but slower. Use create_procurement_search_index only after the user agrees to download and index a bounded package set locally; the tool requires confirm_download=true. Then use search_procurement_index to search downloaded package details with SQLite FTS.

The LLM should start with direct keyword search. If results are weak or the user asks for a broader full-text search, it should explain the tradeoff before creating a local index.

---

ISB Satu Data API (Alternative Data Source)

In addition to direct realtime SPSE/Inaproc queries, PyProc also integrates with the LKPP ISB (Indonesia Satu Data) API. This is a curated, government-maintained dataset that provides:

• Master LPSE — reference list of all LPSE units with their codes and names
• Master K/L/PD — reference list of all government institutions (Kementerian, Lembaga, Perangkat Daerah)
• Tender Umum Publik — public tender data indexed by LPSE code and budget year

The ISB Satu Data API serves as an alternative data source when:

• Direct SPSE/Inaproc queries fail or return errors (e.g., server timeouts, blocked hosts)
• You need reference data (LPSE codes, institution codes) before making SPSE queries
• The user explicitly wants the curated ISB dataset rather than realtime SPSE data

Use the satudata CLI subcommand or the get_master_lpse, get_master_klpd, and get_tender_umum_publik MCP tools to access ISB data.

---

Example LLM Workflows

Here are examples of what you can ask an LLM when PyProc MCP tools are connected:

"Cari data lelang pengadaan laptop pada kementerian keuangan" → LLM uses search_lpse_hosts to resolve "kementerian keuangan" to kemenkeu, then calls search_tender_packages with keywords like ["laptop", "notebook", "komputer"].

"Kalau keyword biasa kurang lengkap, cari full text di detail paket" → LLM explains that local indexing downloads package details, then uses create_procurement_search_index followed by search_procurement_index if the user agrees.

"Find active procurement packages related to cybersecurity in 2025" → LLM uses search_tender_packages with keyword "keamanan siber" or "cybersecurity" across relevant LPSE hosts.

"Summarize tender package 10080116000 on kemenkeu — what's the HPS, who won, and when?" → LLM uses get_tender_detail to retrieve the full package and summarizes key fields.

"Ambil detail untuk 5 paket tender pertama dari hasil pencarian" → LLM uses get_tender_details_bulk with the selected package IDs instead of calling get_tender_detail repeatedly.

"Compare the top 5 infrastructure tenders on pu LPSE by HPS value" → LLM uses search_tender_packages with kategori=PEKERJAAN_KONSTRUKSI, sorts by HPS, and presents a comparison table.

"Generate a vendor research checklist for a company bidding on IT procurement" → LLM analyzes recent IT-related tenders and generates a checklist of requirements, certifications, and typical bid values.

---

Installation

Stable release

pip install pyproc

With MCP support

pip install pyproc[mcp]

Development version

pip install git+https://github.com/wakataw/pyproc.git

Uninstall

pip uninstall pyproc

---

Configuration

The MCP server is configured via environment variables.

Common (all interfaces):

| Variable | Default | Description | |---|---|---| | PYPROC_TIMEOUT | 30 | HTTP request timeout in seconds | | PYPROC_RATE_LIMIT_DELAY | 1.0 | Minimum seconds between requests | | PYPROC_LOG_LEVEL | INFO | Logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL) | | PYPROC_SSL_VERIFY | 0 | Enable SSL/TLS certificate verification (1, true, yes, on to enable) | | PYPROC_MCP_WORKERS | 4 | Worker count for parallel detail fetching |

HTTP interface only:

| Variable | Default | Description | |---|---|---| | PYPROC_MCP_HOST | 0.0.0.0 | HTTP bind address | | PYPROC_MCP_PORT | 8080 | HTTP listen port | | PYPROC_MCP_STATELESS | 0 | Enable stateless mode (no session tracking) | | PYPROC_MCP_JSON_RESPONSE | 1 | Use JSON responses instead of SSE streams | | PYPROC_MCP_SESSION_IDLE_TIMEOUT | 1800 | Session idle timeout in seconds (stateful only) |

---

Cache and Rate Limiting

• MCP server: Does not use persistent caching. Each tool call makes a fresh request to SPSE/Inaproc. Rate limiting enforces a minimum 1-second delay between requests to respect SPSE servers.
• CLI tool: Uses a local SQLite database (.idx file) for download progress tracking and resume support. The cache is disposable per-download.
• Python library: No built-in caching. You control the request lifecycle through the Lpse class methods.

Data freshness depends on SPSE/Inaproc source availability. For the most up-to-date information, visit spse.inaproc.id.

---

Python Library Usage

Initialization

from pyproc import Lpse

lpse = Lpse('kemenkeu', timeout=30)  # timeout in seconds

The Lpse class supports context manager usage:

with Lpse('kemenkeu') as lpse:
    packages = lpse.get_paket_tender(length=10)

Search Tender Packages

from pyproc import Lpse, JenisPengadaan
from pyproc.lpse import By

lpse = Lpse('kemenkeu')

# Basic search
packages = lpse.get_paket_tender(start=0, length=25)

# Search by keyword
packages = lpse.get_paket_tender(search_keyword='sekolah', length=10)

# Filter by category
packages = lpse.get_paket_tender(
    kategori=JenisPengadaan.PENGADAAN_BARANG,
    length=10
)

# Filter by budget year
packages = lpse.get_paket_tender(tahun=2025, length=10)

# Sort by HPS
packages = lpse.get_paket_tender(order=By.HPS, length=10)

# Raw data only (list instead of dict)
data = lpse.get_paket_tender(data_only=True, length=10)

Search Non-Tender Packages

packages = lpse.get_paket_non_tender(start=0, length=25)

Search Pencatatan Packages

Pencatatan non-tender is a separate entity from ordinary non-tender.

packages = lpse.get_paket_pencatatan_non_tender(
    kategori=JenisPengadaan.JASA_LAINNYA,
    rekanan='PT Test',
    tahun=2026,
    instansi_id='L112',
)

swakelola = lpse.get_paket_swakelola(
    tipe_swakelola=1,
    rekanan='Direktorat Arsitektur dan Desain',
    tahun=2026,
    instansi_id='K68',
)

darurat = lpse.get_paket_pengadaan_darurat(
    kategori=JenisPengadaan.PEKERJAAN_KONSTRUKSI,
    tahun=2026,
    instansi_i

…

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [wakataw](https://github.com/wakataw)
- **Source:** [wakataw/pyproc](https://github.com/wakataw/pyproc)
- **License:** MIT
- **Homepage:** https://wakataw.github.io/pyproc/

Install and usage instructions live in the source repository linked above.