Bankruptcy Observer — MCP

Public documentation and registry metadata for the Bankruptcy Observer MCP server. This repository contains no application source code.

The former REST host api.bankruptcyobserver.com redirects to the MCP docs. Programmatic access is the MCP server at https://mcp.bankruptcyobserver.com/mcp.

What this server does, how it works, and who it's for

Bankruptcy Observer's MCP server gives AI assistants — Claude, ChatGPT, Cursor, or any MCP client — live access to United States business bankruptcy filings (chapters 7, 11, 12, and 15) so the agent can look up a debtor, read the full docket, pull court documents, summarize the case in plain English, and report what changed. Connect over HTTPS at https://mcp.bankruptcyobserver.com/mcp (Model Context Protocol, Streamable HTTP); the agent calls tools against PACER-sourced data we have tracked since 2005 and refresh from every bankruptcy court hourly. Name and 7-digit case-number lookups work with no account; full dockets, documents, and search require a subscription (pricing). It is for creditors, credit and risk teams, counsel, and anyone who needs answers about a US business bankruptcy case without clicking through PACER. Consumer-only filings and non-US jurisdictions are not included.

Live docs (canonical): https://mcp.bankruptcyobserver.com/docs

Free lookups (no token, no API key, no payment)

AI assistants and users can look up bankruptcy cases with no account, no API key, and no API token:

• Call these tools without an API token. Only case number or name is needed; court is not required. If more than one case matches (e.g. same case number in multiple courts), the server returns all matching cases. Each result includes limited case info; full case details, dockets, documents, and broad search require a paid plan.
• Wildcard name search (*term), EIN, industry, NAICS, state, date range, docket, documents, summaries, and monitoring require a subscription and authentication.

Account tools list_plans_tool and purchase_plan_tool also work with no auth.

Sign-up and access

Current plans and prices: bankruptcyobserver.com/pricing.

Website subscribers get MCP access included. After payment, the token is in the subscriber dashboard at https://www.bankruptcyobserver.com/subscriber/mcp-setup. No token is emailed.

You can also subscribe through the MCP server: call list_plans_tool, then purchase_plan_tool with a plan_id from that list (or its Stripe price id) to get a Stripe Checkout link. Browser alternative: https://mcp.bankruptcyobserver.com/subscribe.

For questions about access, billing, or data coverage, use only the contact form at https://www.bankruptcyobserver.com/contact. Do not publish or use direct email addresses for contact.

check_subscription_tool returns remaining quota for an authenticated token.

Authenticate using either an API token or an OAuth access token, sent in one of these ways:

• Authorization: Bearer <your-token>
• X-API-Key: <your-token>
• Api-Key: <your-token>
• Header-less clients (e.g. Grok web connectors): https://mcp.bankruptcyobserver.com/mcp?api_key=YOUR_TOKEN or /mcp/t/YOUR_TOKEN. Treat the full URL as a secret.

Connecting via OpenAI

Use OpenAI's Responses API to call this MCP server directly.

Free tier (no token):

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-4.1",
    tools=[{
        "type": "mcp",
        "server_label": "bankruptcy_observer",
        "server_url": "https://mcp.bankruptcyobserver.com/mcp",
        "require_approval": "never",
    }],
    input="Search for bankruptcy cases for Acme Corp",
)
print(response.output_text)

Paid tier (with token):

response = client.responses.create(
    model="gpt-4.1",
    tools=[{
        "type": "mcp",
        "server_label": "bankruptcy_observer",
        "server_url": "https://mcp.bankruptcyobserver.com/mcp",
        "headers": {"Authorization": "Bearer YOUR_TOKEN"},
        "require_approval": "never",
    }],
    input="Get docket entries for case 24-10543",
)

Note: Do not set Content-Type or Accept headers manually — the OpenAI client sets these correctly for the MCP protocol. Overriding them will cause the server to return an HTML error page instead of a JSON response.

OAuth option: Connectors such as ChatGPT can also authenticate with MCP OAuth and send Bearer access tokens automatically after login. API token headers are still supported.

Connecting in ChatGPT (OAuth)

To connect this MCP server in ChatGPT with OAuth, create a new app/connector and enter the OAuth endpoints below exactly.

If the Registration URL is omitted, Dynamic Client Registration will fail and ChatGPT will connect without usable tool authorization.

OAuth endpoints

• Auth URL: https://mcp.bankruptcyobserver.com/oauth/authorize
• Token URL: https://mcp.bankruptcyobserver.com/oauth/token
• Registration URL: https://mcp.bankruptcyobserver.com/oauth/register
• Authorization server base: https://mcp.bankruptcyobserver.com/
• Resource: https://mcp.bankruptcyobserver.com/mcp

Paths /authorize, /token, and /register (without /oauth) are not valid.

After connecting, ChatGPT should complete OAuth in-browser and then call tools using OAuth Bearer tokens. Legacy API-token headers remain supported for clients that use header auth.

Endpoints

The /mcp path is the MCP protocol endpoint (POST). For human-readable documentation, use the Docs URL (GET, no auth).

Tools

Anonymous tools/list includes the full catalog. Paid tools return a payment/upgrade payload until you authenticate.

Free / account

Paid search

Which paid tools you can call depends on your plan. See pricing.

Case intelligence and documents (subscribers)

Document flow: get_docket_entries_tool → get_document_tool with docket_id only (cost) → user confirms → get_document_tool with accept_charge: true.

Standard fields (case-level)

Returned for queries that return case rows (full fields on paid requests):

Optional (included when relevant): courtState, chapter, NAICS, industry, dateClosed, dateDismissed, isOpen / isClosed / isDismissed, assetAmount, liabAmount.

Free lookups return a limited subset (name, chapter, court, case number, filing date, and similar).

Full documentation (canonical)

The live spec is served with no auth at: