MCP API reference

HISAB exposes a local Model Context Protocol (MCP) server. Any MCP-compatible AI client can drive HISAB tools.

Endpoint

HISAB binds the MCP server to http://localhost:7317/mcp. Authentication is bearer-token (token rotates per session, stored at %APPDATA%\HISAB\mcp-token.txt).

Tool surface (excerpt)

list_sheets()                              -> [{ name, used_range, header }]
read_range(sheet, range)                   -> [[...rows...]]
read_values(sheet, addresses)              -> [val, val, ...]
write_cell(sheet, address, value)
write_formula(sheet, address, formula)
bulk_write(sheet, range, values_2d)
execute_analysis(python_code)              -> { stdout, stderr, written_ranges }
ask_user_choice(question, options)         -> selected option
list_skills()                              -> [{ id, name, description }]
read_skill(skill_id)                       -> skill manifest
tool_list() / tool_info(name) / tool_call(...)

JSON-RPC envelope

POST /mcp HTTP/1.1
Authorization: Bearer <token>
Content-Type: application/json

{
  "jsonrpc": "2.0",
  "id": "abc",
  "method": "tools/call",
  "params": {
    "name": "read_range",
    "arguments": { "sheet": "Bank Statement", "range": "A1:E50" }
  }
}

Discovery from Claude clients

HISAB's installer wires the MCP entry into ~/.claude.json automatically. Other MCP clients (Cline, Continue.dev, etc.) can connect manually using the URL + token above.

Note: The MCP server only binds to 127.0.0.1 — never reachable from your LAN or the internet. The bearer token is rotated per session, so even if a token leaks it's invalid after restart.