The MCP servers
Connect Claude Code, Cursor or another MCP client to Bilby to ask it questions, or to this Help Center to look up how Bilby works.
3 min read · Updated 31 Oct 2026 · For developers
Before you begin
Two servers
| Bilby's MCP server | The Help Center's MCP server | |
|---|---|---|
| Address | https://pulse.bilby.ai/api/v1/mcp | https://docs.bilby.ai/mcp |
| What it does | Asks Bilby questions as you, in your organization | Searches and reads these guides |
| Sign-in | Your API key, as a bearer token | None |
| Tools | agent_start, agent_get, agent_update, agent_cancel | search, list_pages, get_page |
Bilby's MCP server
Step 1: Add it to your tool
Bilby's server signs in with an API key in the Authorization header. In Claude Code:
claude mcp add --transport http bilby https://pulse.bilby.ai/api/v1/mcp \
--header "Authorization: Bearer $BILBY_API_KEY"In Cursor, add it to .cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project:
{
"mcpServers": {
"bilby": {
"url": "https://pulse.bilby.ai/api/v1/mcp",
"headers": { "Authorization": "Bearer pulse_…" }
}
}
}Note
The server accepts API keys only. A tool that can connect only by signing in through a browser, with no way to send a header, can't connect to it yet.
Step 2: Ask Bilby from your tool
Ask your tool something that needs Bilby, such as "Ask Bilby what changed in Malaysian data protection rules this year". The tool uses four tools on the server:
| Tool | Arguments | What it does |
|---|---|---|
agent_start | message, and workspaceId for keys with several organizations | Starts a chat with the question. Returns an invocationId, status: "working" and pollAfterMs |
agent_get | invocationId | Returns the chat with its messages, and status: working or completed |
agent_update | invocationId, message | Sends a follow-up in the same chat |
agent_cancel | invocationId | Stops the answer that's running |
agent_start returns before Bilby has answered. The tool then calls agent_get about once a second,
as pollAfterMs suggests, until status is completed. The invocationId is the chat's id, and the
chat appears in Bilby like any other, in the key's organization.
Step 3: Keep it to one organization
With a key for one organization, every chat goes there. With a key for several, the tool must pass a
workspaceId to agent_start. Scoping the key to one organization saves you from that.
The Help Center's MCP server
The same server answers Bilby when you ask it how to do something in Bilby, and the Ask AI box on this site. Adding it to your own tool lets that tool answer from these guides too.
Step 1: Add it to your tool
In Claude Code:
claude mcp add --transport http bilby-help https://docs.bilby.ai/mcpIn Cursor's mcp.json:
{
"mcpServers": {
"bilby-help": { "url": "https://docs.bilby.ai/mcp" }
}
}In the Claude app, add a custom connector under Settings › Connectors with the address
https://docs.bilby.ai/mcp. It needs no sign-in.
Step 2: Ask about Bilby
Ask your tool a how-to question, such as "How do I correct a Radar score in Bilby?". It uses three tools:
| Tool | Arguments | What it does |
|---|---|---|
search | query | Finds the guides that match, best first |
list_pages | none | Lists every guide with its address |
get_page | url, a guide's path such as /radar/radar-scores | Returns the guide as Markdown |
The server only reads. It can't see anyone's Bilby account, so it answers how Bilby works, not what's in your chats.
Tip
Without MCP, the same Markdown is at each guide's address plus .md, such as
https://docs.bilby.ai/radar/radar-scores.md.
Try it in Bilby
Make an API key for Bilby's MCP server.
The Help Center's server needs no key. Add it with the address above.