MCP server

Ownlate exposes a Model Context Protocol server, so an assistant like Claude can read your projects, translate segments, run quality checks and cut releases — with your permission, and only within the workspaces you allow.

Endpoint: https://mcp.ownlate.com

Connecting

The server speaks streamable HTTP and authorises with OAuth 2.1, including dynamic client registration, so most clients need nothing more than the URL.

claude mcp add --transport http ownlate https://mcp.ownlate.com/

The first call returns a 401 with the address of the protected-resource metadata; your client follows it, registers itself, and sends you to a consent screen. There you choose:

Which workspaces the token may reach. It cannot see any other, whatever you are a member of.
Which permissions it may exercise, taken from the same permission codes as everything else.

For a device with no browser, the same server supports the device authorisation flow: the client shows a code, you approve it at platform.ownlate.com/device.

A token can never do more than you can. Granting translation:approve to an assistant when you do not hold it yourself grants nothing.

Configuration files

Claude Desktop — claude_desktop_config.json:

{
  "mcpServers": {
    "ownlate": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.ownlate.com/"]
    }
  }
}

VS Code — .vscode/mcp.json:

{
  "servers": {
    "ownlate": {
      "type": "http",
      "url": "https://mcp.ownlate.com/"
    }
  }
}

What leaves the workspace

Everything a tool returns is read by the model of the client you connected: source texts, translations, keys, glossary terms, comments and the names of the people who wrote them. Connect an assistant only where that is acceptable — for a workspace under an agreement that forbids it, it is not.

What never leaves: the OTA access key of a project, the credentials of any integration, and anything about billing. No tool returns them.

A grant is yours to withdraw: platform.ownlate.com → your profile → connected applications. Withdrawing it stops the tokens at the next call.

What the tools cover

Thirty-two tools, grouped by what they touch.

Getting oriented

Tool	What it returns
`list_workspaces`	The workspaces this token covers
`get_workspace`	One of them
`list_projects`	Projects of a workspace
`get_project`	A project with its source and target languages
`list_languages`	The language catalogue
`add_target_languages`	Adds languages to a project, keeping the existing ones

Reading segments

Tool	What it returns
`list_segments`	Segments of a project, filtered by language and status
`search_segments_by_keys`	Segments for a list of keys; keys with no segment are reported separately
`get_project_progress`	Translated and approved share per language
`get_project_analytics`	Aggregate counts for a project
`list_translation_files`	The files registered in a project
`get_translation_history`	Who changed a translation, when, and what it was before
`export_translations`	A whole locale in one call, shaped like the localisation file

Writing translations

Tool	What it does
`translate_segment`	Saves a translation for one segment in one language
`save_draft_translation`	Stores it as a draft, leaving review to a human
`mark_translation_reviewed`	Records that it was read through
`approve_translation`	Approves it
`reject_translation`	Sends it back for rework
`create_segment`	Adds a key with its source text
`import_segments`	Writes many keys at once — sources without a language, translations with one
`update_segment_source_text`	Rewrites the source string
`set_segment_translatable`	Marks a key that must ship as written

Quality

Tool	What it does
`search_translation_memory`	How similar source texts were translated before, with a score
`list_glossary_entries`	Terms and forbidden wordings
`find_glossary_matches`	The glossary entries occurring in a given text
`check_segment_quality`	Which QA rules a translation violates
`list_segment_comments`	The discussion on a segment
`add_segment_comment`	Leaves a note — a question, or why a wording was chosen

Releasing

Tool	What it does
`list_releases`	Releases with versions, languages and segment counts
`create_release`	Snapshots approved translations into a new release
`recreate_latest_release`	Rebuilds the latest release from what is approved now, keeping its version
`get_ota_distribution`	The project’s distribution — without the access key

What each tool needs

A tool is offered only when the token carries the right, and the call goes through only when you still hold that right in the workspace. Losing a right takes the tool away from the next call on, without reissuing the token.

Right	Tools
`workspace:view`	`list_workspaces`, `get_workspace`, `find_glossary_matches`
`project:view`	`list_projects`, `get_project`, `list_languages`, `list_glossary_entries`, `list_releases`, `list_translation_files`, `get_ota_distribution`
`project:create`	`add_target_languages`, `create_release`, `recreate_latest_release`
`segment:view`	`list_segments`, `search_segments_by_keys`, `get_project_progress`, `get_project_analytics`, `get_translation_history`, `search_translation_memory`, `check_segment_quality`, `list_segment_comments`, `export_translations`
`segment:create`	`create_segment`, `import_segments`, `update_segment_source_text`, `set_segment_translatable`
`translation:write`	`translate_segment`, `save_draft_translation`, `add_segment_comment`
`translation:approve`	`approve_translation`, `reject_translation`, `mark_translation_reviewed`

Every call that changes something is written to the workspace audit log as mcp.<tool>, with the client and the grant it came from.

Resources and prompts

Two resources are readable without calling a tool:

ownlate://project/{projectId}/progress — counts and percentage per language.
ownlate://project/{projectId}/release/latest — version, languages and segment count of the newest release.

Two prompts set up the work an assistant is usually asked for. Both take projectId and language:

translate_missing — walks the untranslated keys, consults the translation memory and the glossary, and writes drafts.
review_translations — reads a language through against the glossary and the quality rules and reports what is off, changing nothing.

Limits

A grant may make 120 calls a minute; beyond that the server answers 429 with Retry-After. A session lives an hour of inactivity, after which the client initialises again on its own.

Working with an assistant

A few patterns this is good at:

Reviewing rather than writing. list_segments with status: needs_review, then check_segment_quality and find_glossary_matches on each — the assistant flags what a human should look at instead of approving on its own.
First-pass translation into a draft. save_draft_translation rather than translate_segment, so nothing enters the review queue unread.
Keeping terminology honest. search_translation_memory before writing, so an assistant reuses the wording your team already approved rather than inventing a synonym.
Cutting a release after a merge. get_project_progress, then create_release when the languages you care about are complete.