MCP servers
Two remote MCP servers. One gives your coding assistant this documentation. The other lets an agent read statements with your API key.
| Server | URL | Authentication | Can |
|---|---|---|---|
| Documentation | https://statementbear.com/api/mcp | None | Search and read these pages. |
| Parsing | https://statementbear.com/api/v1/mcp | Authorization: Bearer sb_... | Read statements, fetch results, create collections. Billed like the API. |
Documentation server#
https://statementbear.com/api/mcpClaude Desktop keeps that JSON in claude_desktop_config.json, Cursor in ~/.cursor/mcp.json, VS Code in .vscode/mcp.json beside the project. Any client that supports a remote MCP server works: it is one HTTP URL with no authentication and no session to manage.
| Tool | What it does |
|---|---|
search_documentation | Finds the pages and headings matching a question, with a summary of each. |
get_documentation_page | Returns a whole page as Markdown, by path or slug. |
list_documentation_pages | Every page, its section and its summary. |
Every page is also exposed as an MCP resource, for clients that attach documents instead of calling tools.
Parsing server#
https://statementbear.com/api/v1/mcpFor an agent that handles statements itself: an assistant reviewing a loan file, a bookkeeping agent, an internal tool. Send your API key in the Authorization header. A test key returns fixtures and bills nothing, so connect one first.
Cursor reads STATEMENTBEAR_KEY from your environment; VS Code asks for the key once and stores it. Neither puts the key in the file, so the file can be committed.
| Tool | Arguments | Returns |
|---|---|---|
parse_statement | file_url (an https URL) or file_base64; optional filename, password, async, and test_scenario on a test key. | The document. Files over 60 pages need async: true, then get_document. |
get_document | id | One document, in any state. |
list_documents | Optional limit (1 to 50, default 10) and status. | A summary of each: id, status, created, filename, pages, issuer and period. |
create_collection | name; optional message and expires_in_days. | The collection, with its upload url. |
parse_statementis billed exactly likePOST /api/v1/documents: $0.08 a page on a live key, from your free pages first.- An error comes back as a tool error: the API's
codeandmessagewhen the API refused the call, or a sentence saying what went wrong when the file could not be fetched. - Documents read through MCP are stored, listed, sent to webhooks and kept for your retention period like any other.
file_urlis downloaded by us, so it must be reachable from the internet. A signed URL from your storage works.
Without MCP#
Every documentation page is served as Markdown at its own URL with a .md suffix. There is an index at /docs/llms.txt and the whole documentation in one file at /docs/llms-full.txt. The Copy page button copies the Markdown of the page you are on.