Loading the setup guide…
HANK · Guides
Connect Claude Code
Connect an agent to the documentation MCP server, then read and publish from your own session. Prefer to work in the browser? Publish a product update.
HANK · Guides
Connect an agent to the documentation MCP server, then read and publish from your own session. Prefer to work in the browser? Publish a product update.
Loading the setup guide…
The documentation platform runs an MCP server at https://newdocs.hank.ai/mcp. Connect an agent to it, and
the agent can read the documentation and write to it.
The server speaks Streamable HTTP. It needs a bearer token on every request.
Tokens are issued in the HANK console at https://console.hank.ai.
hank-docs scope. The server refuses a token without that scope.Keep the key out of your repository. Put it in your shell profile or in a secret manager.
Run this command. Replace YOUR_API_KEY with your key.
claude mcp add hank-docs --transport http https://newdocs.hank.ai/mcp --header "Authorization: Bearer YOUR_API_KEY"To configure the server by file instead, put this block in .mcp.json.
{
"mcpServers": {
"hank-docs": {
"type": "http",
"url": "https://newdocs.hank.ai/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Ask the agent to list the tools. The server answers with names such as search_docs and
upsert_topic. Your client can show them with the server name in front, for example
hank-docs__search_docs, because you gave the server that name in step 2.
A 401 answer means the token is missing, wrong, or does not carry the hank-docs scope.
The server reads your permissions from the token. There are two levels.
Any valid token reads the documentation. It searches pages, reads a page or a section, lists the changelog, reads the API specifications and the glossary, and reads the support tickets that belong to your own organization. It can also write a page, and that write waits for review.
An administrator token publishes. Ask for the administrator permission in the console if you need it. See the Publish a product update guide, which explains that administrator access covers every product, not one.
| Tool | What happens |
|---|---|
upsert_topic | Writes a page. It waits for review, unless an administrator token sends requireReview: false. |
publish_changelog | Publishes a release note at once. It needs an administrator token. |
create_product | Creates a product. It needs an administrator token. |
create_manual | Creates a manual. It needs an administrator token. |
create_chapter | Creates a chapter. It needs an administrator token. |
upload_image | Uploads an image. It needs an administrator token. |
A write from a token without the administrator permission always waits for review. The tool does not fail, and it does not lose your content. Read the returned status to see the result.
The server has NO delete tool. Removing content is a browser task, and it asks you to confirm.
The server checks content before it writes anything. It refuses content that does not build, an image without alt text, a link that does not resolve, and text that looks like patient or personal data. A refusal writes nothing and explains what to correct.
The server also limits writes to 120 requests each minute
for each identity. Above that limit it answers 429 and sends a Retry-After header. Wait for
the time in that header, then try again.
Every task on this page also works in the admin area. See the Publish a product update guide.