Philippine Financial Access MCP Server

A Model Context Protocol server that provides Philippine financial access data to LLMs. 587 BSP-supervised institutions, 37,834 financial access points (bank offices, ATMs, NSSLAs), and coverage analytics with PSGC and 2024 Census population data. Built on Cloudflare Workers with static JSON data.

Public, read-only, no authentication required. Data sourced from the Bangko Sentral ng Pilipinas (BSP) SharePoint API and the 2024 Census of Population via PSGC-MCP. All data bundled at build time for low-latency global access.

Tools

Institution Tools

| Tool | Description | |------|-------------| | search_institutions | Search 587 BSP-supervised institutions by name (fuzzy match on registration and trade name). | | get_institution | Look up an institution by code. Includes contact info, head office address. | | list_institutions | List institutions with optional filters: type, PSGC location, status. | | get_institution_stats | Aggregate counts by type, region, and status. |

Access Point Tools

| Tool | Description | |------|-------------| | search_access_points | Search 37,834 BSP-registered financial access points by institution or location name. Filter by region, province, town, industry, ATM availability. Each record is a single access point. A physical office with multiple ATMs may appear as multiple records. | | get_access_point | Look up a single access point by ID. |

Coverage Analysis Tools

| Tool | Description | |------|-------------| | get_coverage | Coverage report for any PSGC area: access point counts by industry (BANK, ATM ONLY, NSSLA), unique institutions, population ratio. | | find_unbanked_areas | Find municipalities with zero access points. Sorted by population (largest unserved first). | | find_underserved_areas | Rank regions or provinces by population-per-access-point ratio. | | get_institution_footprint | Map an institution's nationwide access point distribution: count by region/province, ATM stats, industry breakdown. | | compare_coverage | Side-by-side coverage comparison of two PSGC areas. |

Industry Types

Access point data includes three industry classifications:

| Industry | Count | Description | |----------|-------|-------------| | BANK | 26,675 | Bank office access points (includes individual ATMs registered at bank offices) | | ATM ONLY | 10,961 | Standalone ATMs not co-located with a bank office | | NSSLA | 198 | Non-Stock Savings and Loan Associations |

Important: Each BSP-registered access point is an individual record. A single physical branch with 3 ATMs may appear as 3-4 records in this dataset. The data measures financial access coverage, not physical branch counts.

Institution Types

| Type | Count | Description | |------|-------|-------------| | universal_commercial | 53 | Universal and Commercial Banks (includes branches of foreign banks, OBUs, representative offices) | | thrift | 42 | Thrift/Savings Banks and Private Development Banks | | rural | 351 | Rural Banks (includes microfinance-oriented rural banks) | | cooperative | 21 | Cooperative Banks | | digital | 6 | Digital Banks | | quasi_bank | 5 | Non-Banks with Quasi-Banking Functions | | non_bank_fi | 109 | Non-Bank Financial Institutions (NSSLAs, financing companies, securities dealers, etc.) |

Coverage Report

The get_coverage tool returns a financial access report for any PSGC area (region, province, or municipality).

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | psgc_code | string | Yes | 10-digit PSGC code | | industry | string | No | Filter by industry type |

Response includes total_access_points, by_industry (BANK/ATM ONLY/NSSLA counts), unique_institutions, with_atm, population, and population_per_access_point.

Institution Footprint

The get_institution_footprint tool maps an institution's nationwide access point distribution using fuzzy name matching.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | institution_name | string | Yes | Institution name to search (e.g. "BDO", "METROBANK") |

Response includes institution_name (canonical), total_access_points, by_region, by_province, with_atm, and by_industry.

Use Cases

Financial Inclusion Analysis

• "Which provinces have the fewest access points per capita?" find_underserved_areas at province level.
• "Which municipalities have zero financial access?" find_unbanked_areas.
• "How does BARMM compare to CALABARZON?" compare_coverage with region PSGC codes. Side-by-side population, access points, and density ratios.

Bank Expansion / Competitive Intelligence

• "Where is BDO vs Metrobank vs BPI?" get_institution_footprint for each. Compare regional concentration and access point counts.
• "What's the competitive landscape in Cavite?" get_coverage for Cavite. Access point counts, unique institutions, breakdown by industry.
• "Which provinces have banking but no NSSLAs?" find_underserved_areas filtered by industry NSSLA.

Fintech Market Sizing

• "Where are ATM-only locations with no bank offices?" search_access_points filtered by industry ATM ONLY in a province, cross-referenced with get_coverage for the industry ratio.
• "Which regions are ATM-heavy but bank-light?" get_coverage across regions. Compare BANK vs ATM ONLY in by_industry. Markets ripe for digital banking.

Policy / Regulatory

• "How many digital banks are operating?" list_institutions with institution_type: "digital". Six licensed.
• "National breakdown of institution types?" get_institution_stats. 587 institutions across 7 categories.
• "Which rural banks serve Mindanao?" list_institutions with region PSGC code + institution_type: "rural".

Real Estate / Municipal Planning

• "Does this municipality have financial access?" get_coverage with its PSGC code. Quick yes/no plus depth of service.
• "How does our town compare to the neighboring one?" compare_coverage. Population-normalized comparison.

Journalism / Data Stories

• "The 10 largest towns with no bank" find_unbanked_areas sorted by population.
• "How concentrated is Philippine banking?" get_institution_footprint for the top 5 banks. BDO alone has 7,479 of 37,834 total access points (19.8%).

Response Format

All data responses are wrapped in a standard envelope:

{
  "_meta": {
    "dataset_version": "1.0.0",
    "dataset_date": "2026-03-12",
    "last_synced": "2026-03-12",
    "source": "Bangko Sentral ng Pilipinas (BSP)",
    "source_url": "https://www.bsp.gov.ph/SitePages/financialstability/Directories.aspx"
  },
  "data": { ... }
}

Access point and coverage tools add population provenance:

{
  "_meta": {
    "...": "...",
    "population_source": "Philippine Statistics Authority (PSA)",
    "population_year": "2024",
    "population_dataset": "2024 Census of Population, PSGC Q4 2025 Publication"
  }
}

Paginated responses add:

{
  "pagination": { "total": 7479, "offset": 0, "limit": 20, "has_more": true }
}

Error responses (isError: true) are returned as plain text JSON without wrapping.

Access Point Record Schema

| Field | Type | Description | |-------|------|-------------| | id | string | Access point identifier | | institution_name | string | Name of the parent institution | | branch_name | string | Access point name (BSP field name) | | industry | string | BANK, ATM ONLY, or NSSLA | | address | string | Street address | | town | string | Municipality/city name | | province | string | Province name | | region | string | Region name | | latitude | number\|null | Latitude (city/district centroid, not address-level) | | longitude | number\|null | Longitude (city/district centroid, not address-level) | | has_atm | boolean | Whether the access point has an ATM | | psgc_muni_code | string | 10-digit PSGC municipality/city code | | region_code | string | 10-digit PSGC region code | | province_code | string | 10-digit PSGC province code |

Institution Record Schema

| Field | Type | Always Present | Description | |-------|------|----------------|-------------| | institution_code | string | Yes | BSP SharePoint list item ID. Stable across ETL runs. | | registration_name | string | Yes | Official BSP registration name | | bank_type | string | Yes | Normalized classification (see types above) | | status | string | Yes | active, closed, under_receivership, or merged | | head_office_address | string | Yes | Head office address as listed by BSP | | psgc_muni_code | string | No | 10-digit PSGC municipality/city code | | region_code | string | No | 10-digit PSGC region code | | province_code | string | No | 10-digit PSGC province code | | contact_person | string | No | President/CEO/Chairman name | | contact_title | string | No | Title (e.g. "President and Chief Executive Officer") | | contact_email | string | No | Contact email address(es) | | contact_phone | string | No | Contact phone number(s) | | website | string | No | Institution website URL | | fax | string | No | Fax number(s) | | num_offices | number | No | Number of offices/branches | | date_sourced | string | Yes | Date the data was fetched (ISO format) | | source_document | string | Yes | Data source identifier |

Connect

Add to your MCP client configuration:

{
  "mcpServers": {
    "ph-financial-access": {
      "url": "https://ph-financial-access.godmode.ph/mcp"
    }
  }
}

Works with Claude Desktop, Cursor, Windsurf, Claude Code, and any MCP-compatible client.

Quick test

curl -X POST https://ph-financial-access.godmode.ph/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "search_access_points",
      "arguments": { "query": "BDO", "limit": 3 }
    }
  }'

Data Sources

| Source | Vintage | Records | Description | |--------|---------|---------|-------------| | BSP Institutions API | March 2026 | 587 | BSP-supervised financial institutions (SharePoint REST API) | | BSP FSAP API | March 2026 | 37,834 | Financial Service Access Points with geocoordinates (SharePoint REST API) | | 2024 Census of Population | Proclamation No. 973 | 1,756 areas | Population counts via PSGC-MCP | | PSGC Q4 2025 Publication | January 13, 2026 | 1,656 municipalities | Geographic codes and hierarchy |

Last synced: March 12, 2026.

Known Data Issues

• Geocoordinate quality. 411 of 37,834 access point records have coordinates outside the Philippines. 167 default to Poughkeepsie, NY (41.60, -73.09) and 101 to Grand Forks, ND (47.93, -97.03), likely geocoding tool defaults. 3 are legitimate overseas access points (BDO Hong Kong). The remaining 140 are scattered globally (geocoding errors). All are nulled out but records preserved for text queries.
• Geocoordinate precision. BSP's geocoding assigns city or district centroids rather than actual addresses. Of 37,423 access points with valid coordinates, only 18,975 unique coordinate pairs exist. 4,917 access points share a coordinate with 10+ others. The worst case is 79 access points at a single point in Calamba, Laguna.
• Access points are not physical branches. BSP registers each financial service access point individually. A single bank branch with 3 ATMs may produce 3-4 records. The data measures financial access coverage density, not physical office counts.
• Manila city district granularity. BSP tags Manila access points with "CITY OF MANILA" as the town, not the individual district (Tondo, Sampaloc, Malate, etc.). This means Manila's 14 city districts (SubMun-level PSGC entities) appear as "unbanked" in find_unbanked_areas even though access points exist in those districts. The access points are matched to the City of Manila entity (1380600000), not to district-level codes.
• Institution PSGC match. 586 of 587 institutions matched to PSGC municipality codes (99.8%). The one unmatched institution is BPI Remittance Center HK, an overseas office with a Hong Kong address.
• Institution codes. Sourced from BSP's SharePoint list item Id field. Stable across ETL runs as long as BSP does not rebuild the list. Not a BSP-issued regulatory identifier.
• Status field. BSP API does not expose closure status. All fetched institutions are marked active. Closed banks are removed from BSP's directory entirely.
• Population is 2024 Census. Point-in-time count, not a live estimate. Fast-growing urban areas may have significantly higher actual population.
• Institution name fragmentation. FSAP data produces 9,623 unique institution name strings, many being the same bank with slight variations (abbreviations, typos). The get_institution_footprint tool uses fuzzy matching to consolidate these, which means a query for "BDO" will also capture ATM installations inside client offices (e.g., "BDO UNIBANK INC-WHITE & CASE GLOBAL OPERATIONS CENTER MANILA LLP"). These are real BSP-registered access points but may surprise users expecting only retail branches.

Related Projects

Part of a suite of Philippine public data MCP servers:

• PSGC MCP -- Philippine geographic codes. 42,000+ entities from barangay to region, with 2024 census population. Use alongside this server for PSGC code lookups.
• LTS MCP -- DHSUD License to Sell verification for Philippine real estate projects.
• PH Holidays MCP -- Philippine holiday calendar

All servers are free, public, and read-only. Data pulled from official Philippine government sources.

Contributing and Issues

Found a data error, a PSGC matching edge case, or an access point with wrong coordinates? Open an issue. The known data issues section above covers the most common ones, but BSP data has its own quirks and the issues list is the best place to track them.

BSP updates their directory periodically. If the data looks stale, open an issue and it will be refreshed.

Data Pipeline

The BSP website renders its directories dynamically from SharePoint lists. We query the SharePoint REST API directly, which returns structured JSON. This approach beats scraping or PDF parsing: we get fields the HTML table doesn't show (fax numbers, office counts, multiple type IDs, geocoordinates).

1. Fetch institutions

node scripts/fetch-bsp-api.js

Fetches 587 institutions from _api/web/lists/getbytitle('Institutions')/items. Classifies each using BSP's three-tier type ID system into 7 normalized bank_type values. Outputs data/banks.json.

2. PSGC join (institutions)

node scripts/psgc-join.js

Fuzzy-matches head office addresses to PSGC municipality codes. Strategy: Metro Manila fast path, candidate extraction from comma-separated address parts, normalization (expand Sta./Sto., strip zip codes), exact match, spelling variants (s/z swap), substring fallback. Result: 586/587 matched (99.8%). Enriches data/banks.json in place.

3. Fetch access points

node scripts/fetch-branches.js

Fetches 37,834 access points from _api/web/lists/getbytitle('FSAP')/items. Splits Title field into institution/access point names. Validates geocoordinates against Philippine bounding box (4.5-21.5 lat, 116-127 lng), nulling 411 out-of-bounds records. Outputs data/branches.json.

4. PSGC join (access points)

node scripts/branch-psgc-join.js

Matches Town/Province/Region text fields to PSGC municipality codes. Four-strategy approach: NCR fast path (11,824), direct normalized match (20,501), province/region disambiguation (5,401), substring fallback (53). Result: 37,779/37,779 in-scope matched (100%). 55 overseas records skipped. Enriches data/branches.json in place.

5. Build population lookup

node scripts/build-population-lookup.js

Extracts 1,756 region/province/city/municipality population records from PSGC-MCP into data/population.json.

6. Verify and deploy

npm test
npm run typecheck
npm run deploy

The ETL is idempotent. Each run overwrites the previous data files. Review the diff before deploying.

Requires PSGC-MCP as a sibling directory for steps 2, 4, and 5. Not needed for development if data/*.json files already exist.

Check for updates

./scripts/check-updates.sh

Checks if BSP has updated their directory since the last sync.

Development

git clone https://github.com/GodModeArch/ph-financial-access-mcp.git
cd ph-financial-access-mcp
npm install
npm run dev

Dev server starts at http://localhost:8787. Connect your MCP client to http://localhost:8787/mcp.

npm test             # Run tests
npm run typecheck    # TypeScript strict mode check

Project Structure

src/
  index.ts              # Cloudflare Worker entry, McpAgent class
  tools.ts              # 11 MCP tool definitions with Zod schemas
  data.ts               # Institution search, filter, stats functions
  access-point-data.ts  # Access point search, coverage, footprint functions
  response.ts           # Response envelope helpers
  types.ts              # TypeScript interfaces
data/
  banks.json            # 587 institutions (bundled into worker)
  branches.json         # 37,834 access points (bundled into worker)
  population.json       # 2024 Census population by area (1,756 records)
scripts/
  fetch-bsp-api.js      # Institution ETL from BSP SharePoint API
  fetch-branches.js     # Access point ETL from BSP FSAP SharePoint list
  psgc-join.js          # Institution address -> PSGC code matching
  branch-psgc-join.js   # Access point town/province -> PSGC code matching
  build-population-lookup.js  # Population data extraction
  check-updates.sh      # BSP update checker
test/
  data.test.ts              # Institution function tests
  access-point-data.test.ts # Access point function tests

Architecture

Runs on Cloudflare Workers with Durable Objects. All data is bundled as static JSON at build time (2.5MB gzip total). No database, no external API calls at runtime. Queries execute in-memory against the bundled dataset.

Built by

Aaron Zara -- Fractional CTO at Godmode Digital

This MCP came out of needing structured, queryable Philippine banking data for AI agents. The BSP publishes all of this data, but not in a format machines can easily consume. We wrote the ETL pipeline, PSGC geographic enrichment, and queryable API layer. The data is public. The code is the recipe, the hosted instance is the restaurant.

For enterprise SLAs, custom integrations, or other PH data sources: godmode.ph

License

MIT