Two-stage classifier for Japanese tax accounting. Stage 1: keyword dictionary match (14 categories × ~50 keywords). Stage 2 (= deferred to Phase 1.B): Claude API fallback. Returns 勘定科目 + 税区分 + confidence.
AI agents call classify_transaction to retrieve information from Cockpit without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
classify_transaction performs read-only analysis: it queries transaction data against a keyword dictionary and runs inference to assign Japanese accounting codes (勘定科目), tax categories (税区分), and confidence scores. It produces outputs but does not create, modify, delete, or execute external operations. The two-stage matching process is deterministic analysis with no side effects on the transaction record itself.
From the tool's definition Returns 勘定科目 + 税区分 + confidence—outputs classification results without modifying transaction data. Described as a 'classifier' using 'keyword dictionary match' and 'Claude API fallback' for analysis, not creation or alteration of records.
Attacks that exploit this kind of access
Two-stage classifier for Japanese tax accounting. Stage 1: keyword dictionary match (14 categories × ~50 keywords). Stage 2 (= deferred to Phase 1.B): Claude API fallback. Returns 勘定科目 + 税区分 + confidence. It is categorised as a Read tool in the Cockpit MCP Server, which means it retrieves data without modifying state.
Register the Cockpit MCP server in PolicyLayer and add a rule for classify_transaction: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Cockpit. Nothing to install.
classify_transaction is a Read tool with low risk. Read-only tools are generally safe to allow by default.
Yes. Add a rate_limit block to the classify_transaction rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for classify_transaction. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
classify_transaction is provided by the Cockpit MCP server (michielinksee/kansei-link-cockpit). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Every MCP server has a record like this.
Type a name, get the same breakdown: verified identity, auth posture, risk grade, capabilities, recommended policy.
Teams ship this data inside their own products. See what a licence covers →