Use Comma with Cursor
Cursor supports the Model Context Protocol natively. Add Comma as an MCP
server in your Cursor config and the agent picks up the full set of report
and comment tools — same scoped token shape as the REST API.
One-click install
Cursor supports install deeplinks — clicking this from a machine with
Cursor installed opens the MCP install prompt with Comma pre-filled:
Add Comma to Cursor →
The deeplink ships with a placeholder credential. After installing, open
Cursor Settings → MCP → comma and replace YOUR_COMMA_TOKEN in the
Authorization header with a real token from
Settings → API tokens → New token on Comma — the server won't
authenticate until you do.
Prefer to see exactly what lands in your config? Use the manual path
below — it's the same entry.
Add Comma to Cursor manually
Open your Cursor MCP configuration. Most setups use one of:
- Global:
~/.cursor/mcp.json
- Per workspace:
.cursor/mcp.json in the project root
Add a Comma entry:
{
"mcpServers": {
"comma": {
"url": "https://commareports.com/api/mcp",
"headers": {
"Authorization": "Bearer comma_sk_…"
}
}
}
}
Replace comma_sk_… with a token from
Settings → API tokens → New token on Comma. Scope it to the actions you
want the agent to take (typically reports:write and comments:write).
Restart Cursor. The agent now sees the Comma tools.
What the agent can do
Once Comma is attached, the agent has the full MCP toolset:
| Capability |
Tools |
| Publish a report |
create_report, update_report |
| Find existing reports |
list_reports, search_reports, get_report, read_report |
| Leave / reply to comments |
add_comment, reply_to_comment, set_comment_status, list_comments |
| Request review |
request_review |
| Manage assets |
upload_report_asset, list_report_assets, delete_report_asset |
The agent picks the right tool from its reasoning — there are no
Cursor-specific commands to memorize.
Common workflows
Publish at the end of a task
You ask Cursor to "summarize this PR's changes and publish a release note
to Comma." The agent:
- Reads the diff and your codebase context.
- Generates HTML for the release note.
- Calls
create_report to publish.
- Returns the share link.
Pull context from prior reports
You ask Cursor to "look at last month's eval results and update the failing
prompts." The agent:
- Calls
search_reports("eval results") to find the latest one.
- Calls
read_report to load the HTML.
- Reasons about which prompts are failing.
- Edits the prompts in your repo.
- Optionally calls
add_comment on the eval report to record the change.
Carry on a thread
You ask Cursor to "address the comments left on last week's analytics
report." The agent:
- Calls
list_comments on the report.
- Reads the open threads, including the anchored text.
- Makes the requested changes (in code, in data, in copy).
- Calls
reply_to_comment to summarize what changed.
- Optionally calls
set_comment_status to mark threads resolved.
Token shape
The token in your Cursor config:
- Lives in
~/.cursor/mcp.json (or workspace .cursor/mcp.json). Treat it
like an API key: avoid committing the per-workspace file if your repo is
public.
- Is scoped — you choose which actions it can take when you create it on
Comma.
- Is revocable in one click from Settings → API tokens. Revocation
applies immediately to the REST API, the MCP server, and any routines
the token created.
- Is rate-limited per token.
For an extra layer, create separate tokens per project — one per workspace
.cursor/mcp.json — and revoke individually when a project ends.
The deliberate scope
Comma's MCP surface is intentionally narrow:
- Reports, comments, asset management, review requests.
- No code execution on Comma's side; the agent runs on your machine.
- No data layer; the agent's skill fetches its own data.
- No cross-posting to Slack/email/GitHub; Comma posts to a Comma report.
If you want refresh-on-a-cron behavior alongside the on-demand MCP surface,
attach a routine — see Routines →.
Try it
Drop the snippet into mcp.json, generate a token, restart Cursor, and ask
the agent to publish anything you have lying around as HTML. From the web
UI, leave a comment and ask the agent what it sees on its next turn.
Generate a token →
Related