Bareun — Korean NLP & Spell/Grammar Checking
Korean NLP MCP server: morphological analysis, tokenization, spell & grammar checking (Bareun)
Install
mcp_config.json
{
"mcpServers": {
"ai-bareun-bareun": {
"url": "https://api.bareun.ai/mcp",
"type": "streamable-http"
}
}
}Documentation
Bareun MCP Server — Korean NLP & Spell/Grammar Checking
바른(Bareun) is a Korean natural-language platform. This is its MCP (Model Context Protocol) server — it lets any MCP-compatible AI tool (Claude, Cursor, VS Code, Claude Desktop, …) perform Korean morphological analysis and spell/grammar correction by calling Bareun as a tool.
Large language models still miss the subtle spacing, particle agreement, and confusable-word rules of Korean. Plug Bareun in as an MCP tool and your agent can hand off analysis and proofreading to a dedicated Korean engine, then use the result to produce more accurate Korean output.
- Hosted endpoint:
https://api.bareun.ai/mcp - Transport: Streamable HTTP (JSON-RPC 2.0) — no SSE, no extra port, no install
- Auth: API key (
api-keyheader orAuthorization: Bearer <key>) - Get an API key: https://bareun.ai
The
/mcpendpoint is available on the spell-checker–included build of Bareun (the morphological-analysis-only build does not expose/mcp). The same endpoint works on self-hosted/on-prem installs — just swap the host.
What it looks like
Spelling & spacing — correct_grammar (real output from https://api.bareun.ai/mcp):
| In | 회의결과를 정리해서 내일까지 보내주시기 바람니다. |
| Out | 회의 결과를 정리해서 내일까지 보내 주시기 바랍니다. |
Every fix comes back as a block, so an agent can explain the edit instead of silently rewriting the sentence:
| Original | Corrected | Category | Rule |
|---|---|---|---|
| 회의결과를 | 회의 결과를 | SPACING | compound noun spacing |
| 보내주시기 | 보내 주시기 | SPACING | auxiliary-verb spacing |
| 바람니다. | 바랍니다. | TYPO | misspelling |
Morphological analysis — analyze_syntax (format: compact):
| In | 나는 학교에 간다. |
| Out | 나/NP 는/JX 학교/NNG 에/JKB 가/VV ㄴ다/EF ./SF |
Homograph senses — analyze_syntax with with_sense: true (beta): 배 in 배가 아프다
comes back with senseNo: 1 and its dictionary definition — 사람이나 동물의 몸에서 …
가슴과 엉덩이 사이의 부위 (belly, probability 0.87), plus the Urimalsaem entry id — so the
agent knows which 배 it is reading.
Tools
| Tool | What it does | Key inputs |
|---|---|---|
analyze_syntax | Splits a sentence into words/morphemes and tags parts of speech (morphological analysis). | text (required), auto_split_sentence, auto_spacing, auto_jointing, custom_dict_names, encoding, format (full|compact), with_sense |
analyze_syntax_raw | Same analysis without post-processing (no compound-noun/verb splitting, no auto spacing, no custom dictionaries) — the raw model output. | text (required), auto_split_sentence, encoding, format, with_sense |
search_dict | Searches the Urimalsaem Korean dictionary (~1.1M entries) by jamo (phoneme-level) slot patterns — conditions like "verbs whose stem ends in the ㅎ coda" or "words ending in -아지" that cannot be expressed with composed Hangul syllables. | pattern (required), anchor (word|prefix|suffix|contains), pos, std_only, with_definition, limit, count_only |
tokenize | Splits a sentence into word (token) units. | text (required), auto_spacing, encoding |
correct_grammar | Corrects spelling/spacing and returns correction blocks. | text (required), custom_dict_names, + 9 boolean correction options |
list_pos_tags | Returns the 47 part-of-speech tags Bareun uses (code · name · class). | (none) |
correct_grammar options (all boolean, default off): treat_as_title,
disable_split_sentence, disable_caret_spacing, disable_vx_spacing,
enable_limited_punctuation, disable_confusion, enable_cleanup_whitespace,
disable_typo_correction, enable_sentence_check.
encoding controls the unit for morpheme offsets: utf32 (default, code points
— matches Python), utf16 (JS/Java), utf8 (bytes — Go/C++).
with_sense — homograph sense disambiguation (WSD, beta). Korean writes many
unrelated words identically: 배 can be belly, ship, or pear. Set with_sense: true
on analyze_syntax/analyze_syntax_raw and each content morpheme carries a sense
object — the dictionary sense number, its Korean definition, the Urimalsaem entry id, and
the probability of the chosen sense among that word's candidate senses. It is off by
default (one extra model pass; responses are byte-identical to before when omitted),
and format=compact renders it inline as 배__002/NNG. This feature is in beta and
ships officially with Bareun 3.1.0.
Resources
| Resource URI | Contents | Auth |
|---|---|---|
bareun://pos-tags | The 47 POS tags (code · name · class) — same data as list_pos_tags | API key |
bareun://server-info | Server metadata — name · version · build · active tools/resources | API key |
bareun://custom-dicts | Names of custom-dictionary domains registered for the key | valid API key |
Quick start
Tip — register globally. Most tools default to project scope (the server is only available in one project). To use Bareun across all your projects, register it at global / user scope as shown below.
Claude Code
# -s user → global: available in every project
claude mcp add -s user --transport http bareun https://api.bareun.ai/mcp \
--header "api-key: YOUR_API_KEY"
Omit -s user for project-local scope. Check with claude mcp get bareun.
Cursor
Global: ~/.cursor/mcp.json · Project: <project>/.cursor/mcp.json
{
"mcpServers": {
"bareun": {
"url": "https://api.bareun.ai/mcp",
"headers": { "api-key": "YOUR_API_KEY" }
}
}
}
VS Code
Global: run MCP: Open User Configuration · Project: <project>/.vscode/mcp.json
{
"servers": {
"bareun": {
"type": "http",
"url": "https://api.bareun.ai/mcp",
"headers": { "api-key": "YOUR_API_KEY" }
}
}
}
Claude Desktop — claude_desktop_config.json
Claude Desktop bridges header-authenticated remote servers via mcp-remote
(Node.js required):
{
"mcpServers": {
"bareun": {
"command": "npx",
"args": [
"-y", "mcp-remote",
"https://api.bareun.ai/mcp",
"--header", "api-key: YOUR_API_KEY"
]
}
}
}
Cline — cline_mcp_settings.json
Open MCP Servers → Configure MCP Servers in Cline, then add:
{
"mcpServers": {
"bareun": {
"type": "streamableHttp",
"url": "https://api.bareun.ai/mcp",
"headers": { "api-key": "YOUR_API_KEY" },
"disabled": false,
"autoApprove": []
}
}
}
All five tools are read-only (readOnlyHint), so listing them in autoApprove is
safe if you would rather not confirm every call.
ChatGPT — Developer Mode
ChatGPT talks to Streamable HTTP servers directly — no directory review needed. Turn on Developer mode (Settings → Apps & Connectors → Advanced), then Create a connector:
- URL:
https://api.bareun.ai/mcp - Authentication: OAuth — Bareun runs its own OAuth 2.1 (PKCE) endpoint, so
ChatGPT opens a login page where you paste your Bareun API key. (ChatGPT's connector
dialog offers OAuth or no-auth; if your build also lets you set request headers, an
api-keyheader works just as well.)
Menu wording and plan availability shift between ChatGPT releases — Developer mode is an account-level toggle, and on Business/Enterprise a workspace owner enables it first.
Test with MCP Inspector
npx @modelcontextprotocol/inspector
Set Transport to Streamable HTTP, URL to https://api.bareun.ai/mcp, and
add header api-key: YOUR_API_KEY.
Example
// tools/call → analyze_syntax (format: compact)
{ "text": "나는 학교에 간다.", "format": "compact" }
// → "나/NP 는/JX 학교/NNG 에/JKB 가/VV ㄴ다/EF ./SF"
Links
- Service: https://bareun.ai
- Docs: https://bareun.ai/docs
- MCP guide: https://bareun.ai/docs/howtouse/mcp
- API keys & usage: https://bareun.ai/docs/howtouse/cloud-api
License
The contents of this repository (documentation, registry manifests, examples) are released under the MIT License. The Bareun engine itself is a proprietary service operated by Baikal AI; access is governed by the bareun.ai terms of service.
Sourced from the repository README.
More in AI & Agents
- PonytailMakes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.109,599
- AgentsMulti-harness agentic plugin marketplace for Claude Code, Codex, Cursor, OpenCode, GitHub Copilot, and Google Antigravity39,079
- Frontend SlidesCreate beautiful slides on the web using a coding agent's frontend skills28,060
- Agent Skills Search ServerSearch and discover Agent Skills from the skills.sh registry. Powered by HAPI MCP server.24,658
- Agency Agents Zh🎭 267 个即插即用的 AI 专家角色 — 支持 Hermes Agent/Claude Code/Cursor/Copilot 等 18 种工具,覆盖工程/设计/营销/金融等 20 个部门。含 52 个中国市场原创智能体(小红书/抖音/微信/飞书/钉钉等)。搭配编排器 agency-orchestrator,一句话即可让多位专家按 DAG 自动协作。19,868
- Watermarks RemoverStrip multi-vendor AI provenance marks: Unicode text hygiene, statistical rewrite hooks, and C2PA/metadata from PNG/JPEG/SVG/PDF/DOCX/HTML/MD17,822