Japanese Language MCP — JMDict and grammar for AI | yomeru.ai

Connect ChatGPT, Claude, or Codex to JMDict, Tatoeba, and Grammar Lab via a free unauthenticated MCP endpoint at yomeru.ai/api/mcp.

The Japanese Language MCP is a public Model Context Protocol server that exposes yomeru.ai's premium Japanese language tools to any MCP-compatible AI client. It returns structured JSON — JMDict-backed dictionary cards, full grammar breakdowns, example sentences with source attribution — so your AI agent can reason about Japanese with the same data the website uses.

Endpoint: https://yomeru.ai/api/mcp Transport: Streamable HTTP Authentication: None (anonymous, per-IP rate limited)

See the marketing overview & live setup tabs →

Tools exposed

The MCP advertises four tools. Each is callable from your AI client by name.

grammar_lab

Full Japanese sentence breakdown — grammar points, vocab callouts, JLPT levels, markdown output.

grammar_lab_fast

Same Grammar Lab analysis on smaller AI models. Trades a touch of depth for sub-second responses.

word_lab

Complete dictionary cards from JMDict (~200k entries) + JMnedict + Tatoeba. AI fallback for words outside JMDict.

example_sentences

1–5 example sentences per word, each tagged with its source (tatoeba, jmdict, ai_generated).

Setup — pick your client

The marketing page at /what-is-mcp walks through each path with click-to-copy values.

ChatGPT (Custom MCP)

Open the Custom MCP → New App dialog and fill in:

Name: Yomeru.ai Japanese MCP
Description: Lookup Japanese data
MCP Server URL: https://yomeru.ai/api/mcp
Authentication: No Auth

Acknowledge the custom-MCP risk warning to enable the connector.

Claude (Add custom connector)

Settings → Connectors → Add custom connector:

Name: Yomeru.ai Japanese MCP
URL: https://yomeru.ai/api/mcp

Click Add. Anthropic will surface a "use connectors only from developers you trust" notice.

Claude Code (CLI)

claude mcp add --transport http yomeru https://yomeru.ai/api/mcp

The four tools appear in your next Claude Code session.

Codex (CLI)

codex mcp add yomeru --url https://yomeru.ai/api/mcp

Verify with codex mcp list.

Rate limits

Per-IP, aggregated by IPv6 /64. Strong by design — the surface is public and unauthenticated.

Tool	per minute	per hour	per day
`grammar_lab` (LLM)	1	2	5
`grammar_lab_fast`	1	10	30
`word_lab`	2	20	60
`example_sentences` (LLM)	3	30	100

A cluster-wide LLM cost circuit breaker bounds total spend; on trip, generating tools return SYSTEM_BUSY with a retryAfterSec hint.

Need higher limits? Sign up — authenticated users get account-level limits via the website.

Data sources

JMDict (~200k entries) — canonical Japanese-English dictionary, community-curated.
JMnedict — Japanese names dictionary for proper nouns.
Tatoeba — bilingual example sentences.
Curated grammar resource set — JLPT N5 → N1+ explanations, used by Grammar Lab.
AI fallback layer — for words outside JMDict and the 5th example sentence. AI-generated content is tagged source: "ai_generated" and status: "pending" until reviewed.

Output shapes

Every row in every response carries a source field (jmdict, jmnedict, tatoeba, ai_generated) and a status field (approved | pending). MCP-driven AI generations land in staging tables and are not promoted to canonical until reviewed — your client can branch on status to warn users about pending data.

For the full per-tool input and output schemas, see Tool reference.

Privacy & safety

No raw IPs stored. All abuse-tracking uses salted SHA-256 hashes.
No per-row search history for MCP traffic — aggregate counters only.
MCP-generated content is quarantined to staging tables until human-reviewed; your queries cannot poison the canonical website corpus.
Inputs are NFKC-normalized + invisible-character-stripped before validation; rejects pure-romaji, control characters, or sub-30%-Japanese inputs.

Japanese Language MCP — JMDict and grammar for AI | yomeru.ai

grammar_lab

grammar_lab_fast

word_lab

example_sentences

On this page