Stark County CJIS Court Data Puller

What this skill does

Pulls civil case filings from the Stark County CJIS portal at https://www.starkcjis.org. Fetches case details including property addresses, then builds a multi-sheet Excel workbook optimized for real estate wholesaling lead generation.

Before you start — ask the user

1. What case type? Default: foreclosures. Other options: probates, tax liens, general civil.
2. What date range?
   • Today only (default — use this for "daily pull" or "run the pull")
   • Last N days
   • Specific month
   • Full year-to-date
3. Is starkcjis.org already open in Chrome? If not, navigate there first.

---

Critical technical constraint

Python cannot reach CJIS APIs directly — the sandbox VM's outbound requests are blocked (ProxyError). All API calls must be made via JavaScript injected into the user's Chrome browser using the javascript_tool. Never attempt requests.get() or urllib against starkcjis.org.

---

Step 1: Navigate to the portal

Make sure Chrome is on:

https://www.starkcjis.org/#/search

---

Step 2: Query the Search API

Stark County CJIS Advanced Search endpoint:

GET /api/search/advanced?criteria={JSON}&isDocket=true

⚠️ Hard cap of 1,000 rows per query. If pulling more than ~2–3 weeks of foreclosures, chunk by month (one query per month) and merge results.

Building the criteria

// Standard foreclosure query — today only (daily pull default)
const today = new Date().toISOString().split('T')[0]; // e.g. "2026-04-14"
const criteria = JSON.stringify({
  "scCode": "E",   // 'E' = foreclosure in CPC (Common Pleas); 'F' = foreclosure in municipal courts
  "dateRange": { "startDate": today, "endDate": today }
});
const resp = await fetch(`/api/search/advanced?criteria=${encodeURIComponent(criteria)}&isDocket=true`);
const data = await resp.json();
console.log('Total:', data.totalCount, '| Returned:', data.cases?.length);

Known Stark County scCodes (verified)

Case Type	Court	scCode
Foreclosure	CPC (Common Pleas)	E
Foreclosure	Municipal (CMC/MMC/AMC)	F

For an unfamiliar case type, verify the scCode first by looking up a known case:

const r = await fetch('/api/search/advanced?criteria=' + 
  encodeURIComponent(JSON.stringify({"caseNumber": "2026CV00741"})) + '&isDocket=true');
const d = await r.json();
console.log(d.cases?.[0]?.scCode); // confirms "E" for CPC foreclosure

Court filtering

To get all courts (CPC + all municipal), run two queries and merge:

1. CPC: add "court": "CPC" to criteria
2. Municipal: use scCode: "F" without court filter

Or omit scCode entirely and filter client-side:

• CPC case numbers match: 2026CVnnnnn (no F)
• Municipal case numbers match: 2026CVFnnnnn

Chunking by month for large date ranges

async function fetchMonth(year, month) {
  const start = `${year}-${String(month).padStart(2,'0')}-01`;
  const end = new Date(year, month, 0).toISOString().split('T')[0];
  const criteria = JSON.stringify({
    "scCode": "E",
    "dateRange": { "startDate": start, "endDate": end }
  });
  const r = await fetch(`/api/search/advanced?criteria=${encodeURIComponent(criteria)}&isDocket=true`);
  const d = await r.json();
  return d.cases || [];
}
// Example: pull Jan–Apr 2026
const allCases = (await Promise.all([1,2,3,4].map(m => fetchMonth(2026, m)))).flat();

---

Step 3: Extract case data from results

For each case in data.cases, capture:

• _id — MongoDB ObjectId (needed for address lookup in Step 4)
• caseNumber — e.g. 2026CV00741 or 2026CVF00277
• fileDate — ISO date string
• court.code — CPC, CMC, MMC, AMC, etc.
• parties[0].name — plaintiff / lender
• parties[1].name — defendant / homeowner

Build a map: { caseNumber → _id } for the address fetch step.

---

Step 4: Fetch property addresses

The search API does NOT include addresses. Hit the case detail endpoint for each case:

GET /api/case/{mongoObjectId}

The defendant's participants[].address is the property address. Find the defendant participant (role contains "defendant"), or fall back to participants[1].

Batch 10 at a time:

// Store results globally so they survive across JS tool calls
window._addrMap = {};

async function fetchAddressBatch(entries) {  // entries = [[caseNum, mongoId], ...]
  await Promise.all(entries.map(async ([cn, id]) => {
    const r = await fetch(`/api/case/${id}`);
    const d = await r.json();
    const def = d.participants?.find(p =>
      p.role?.toLowerCase().includes('def')
    ) || d.participants?.[1];
    const a = def?.address;
    window._addrMap[cn] = a
      ? [a.line1, a.city, a.state, a.zip].filter(Boolean).join(', ')
      : 'ADDRESS UNKNOWN';
  }));
}

// Call in a loop, 10 at a time:
const entries = Object.entries(caseIdMap);
for (let i = 0; i < entries.length; i += 10) {
  await fetchAddressBatch(entries.slice(i, i + 10));
}
Object.keys(window._addrMap).length; // confirm count

---

Step 5: Transfer data from browser to Python

⚠️ The JS tool truncates output at ~1,400 characters. Never return a large JSON object directly — it will be silently cut off. Always use the chunked transfer pattern.

// Encode all case data as pipe-delimited lines
const lines = window._cases.map(c =>
  [c.caseNumber, c.fileDate, c.court, c.plaintiff, c.defendant,
   window._addrMap[c.caseNumber] || ''].join('|')
).join('\n');

// Break into ~800-char chunks stored in a global array
window._chunks = [];
for (let i = 0; i < lines.length; i += 800) {
  window._chunks.push(lines.slice(i, i + 800));
}
window._chunks.length; // returns chunk count — read this first

Then read chunks one at a time in separate JS tool calls:

window._chunks[0]   // chunk 0
window._chunks[1]   // chunk 1
// ... up to window._chunks[N-1]

Parse in Python:

all_lines = []
for chunk in chunks:        # list of strings you read from browser
    all_lines.extend(chunk.strip().split('\n'))

cases = []
for line in all_lines:
    if '|' not in line:
        continue
    case_num, file_date, court, plaintiff, defendant, address = line.split('|', 5)
    cases.append({...})

---

Step 6: Build the Excel workbook

Use openpyxl. Four-sheet structure:

Sheet	Contents
All Cases	Every case: Case #, Court, Full Court Name, File Date, Type, Plaintiff/Lender, Defendant/Homeowner, Property Address, CJIS Link
By Lender	Ranked plaintiff counts — useful for spotting the most active servicers/banks
Monthly Trend	Filing counts by month (Total / Mortgage / Tax)
CPC Mortgage Leads	CPC cases only, filtered to mortgage/lender types — the highest-value lead list

Classify lender type (for color coding and the Leads sheet filter):

def lender_type(name):
    n = name.upper()
    if any(x in n for x in ['TREASURER', 'TAX EASE']):
        return 'Tax Foreclosure'
    if any(x in n for x in ['BANK', 'MORTGAGE', 'SERVICING', 'LOAN', 'LENDING',
                              'FINANCE', 'FUND', 'SAVINGS', 'CREDIT UNION']):
        return 'Mortgage / Lender'
    return 'Foreclosure'

Style constants:

HDR   = PatternFill('solid', start_color='1F4E79')   # dark blue header
TAX   = PatternFill('solid', start_color='FFF2CC')   # yellow = tax foreclosure
MTGE  = PatternFill('solid', start_color='E2EFDA')   # green = mortgage
WHITE = PatternFill('solid', start_color='FFFFFF')

Save location

# Always save to the user's mounted Claude folder
out = '/sessions/.../mnt/Claude/Stark_County_Foreclosures_2026.xlsx'
# For daily snapshots: Stark_County_Foreclosures_2026-04-14.xlsx

---

Step 7: Present the file

Use mcp__cowork__present_files after saving. Report:

• Total cases found and date range
• How many have property addresses
• How many landed in the CPC Mortgage Leads sheet

---

Daily pull behavior

When the user says "run the daily pull" or "pull today's filings":

• Date range = today only
• File name = Stark_County_Foreclosures_{YYYY-MM-DD}.xlsx
• If zero cases were filed today (common on weekends/holidays), say so clearly rather than producing an empty file

For ongoing use, consider keeping a master YTD file that new daily records get appended to (de-duplicate by case number). Ask the user which they prefer.

---

Troubleshooting

Problem	Fix
totalCount is 0	Verify scCode by looking up a known case (see Step 2). Try without scCode to see what's returned.
Results capped at exactly 1,000	Chunk by month — you're hitting the row cap
JS output appears cut off	You're not using the chunked pattern — store in window._chunks and read sequentially
Address is a servicer's office address	Normal for some cases — include as-is
ADDRESS UNKNOWN	Filed without address (common for estates/unknown heirs) — include as-is
Python ProxyError on CJIS URLs	Expected — you cannot call CJIS from Python. JS in Chrome only.