> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sazabi.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Connect a coding agent

> Add the cve.now MCP server to Claude Code, Codex, OpenCode, Cursor, or any agent that supports remote MCP, so it can look up vulnerabilities, read the feed, and check your lockfile. No login or API key.

cve.now runs a public MCP server at `https://mcp.cve.now/mcp`. It needs no login, OAuth, or API key. Once connected, your agent can look up any vulnerability, read the latest feed, and check the packages in your project against the catalog.

## Prerequisites

* A coding agent that supports remote MCP servers over Streamable HTTP, such as Claude Code, Codex, OpenCode, or Cursor.

## Add the server

> [tab] Claude Code
>
> Run this in your terminal:
>
> ```bash theme={"system"}
> claude mcp add --transport http cve-now https://mcp.cve.now/mcp
> ```

> [tab] Codex
>
> Run this in your terminal. It also configures the Codex editor extension:
>
> ```bash theme={"system"}
> codex mcp add cve-now --url https://mcp.cve.now/mcp
> ```

> [tab] OpenCode
>
> Add the server to `opencode.json` in your project, or to `~/.config/opencode/opencode.json` for every project:
>
> ```json theme={"system"}
> {
>   "$schema": "https://opencode.ai/config.json",
>   "mcp": {
>     "cve-now": {
>       "type": "remote",
>       "url": "https://mcp.cve.now/mcp",
>       "enabled": true
>     }
>   }
> }
> ```

> [tab] Cursor
>
> Add the server to `.cursor/mcp.json` in your project, then restart Cursor:
>
> ```json theme={"system"}
> {
>   "mcpServers": {
>     "cve-now": {
>       "url": "https://mcp.cve.now/mcp"
>     }
>   }
> }
> ```

> [tab] Other agents
>
> Add `https://mcp.cve.now/mcp` as a Streamable HTTP MCP server in your agent's settings.

The [cve.now homepage](https://cve.now/#mcp) also has a copyable prompt that tells your agent to install the server and check your project.

## Tell your agent when to use it

A connected server only helps if the agent reaches for it. Add this block to your project's `AGENTS.md` (Codex, OpenCode, Cursor) or `CLAUDE.md` (Claude Code), so dependency questions go through cve.now:

```md theme={"system"}
## Dependency vulnerabilities

Use the cve.now MCP server (https://mcp.cve.now/mcp, no API key) for vulnerability checks:

- Before adding or upgrading a dependency, and when asked whether this project is affected by a CVE, run the `stack.check` operation with the lockfile's contents (or the packages and exact versions).
- For any CVE, GHSA or OSV id, run `cve.get` to see whether it is in CISA KEV, whether its EPSS score has reached 10% (and when), and the fixed versions.
- No matches means cve.now found none, not that the packages are safe.
```

## Tools and operations

The server has two tools that never change: `search` finds an operation and `execute` runs one. Call `search` with no arguments to list every operation, with a query to narrow them, or with `includeInputSchema: true` and an operation's name for its exact arguments. Run one with `execute`: `{"name": "cve.get", "arguments": {"id": "CVE-2021-44228"}}`. The data search is the operation `entities.search`, not the `search` tool.

| Operation | What it does | Inputs |
| - | - | - |
| `cve.get` | Returns one entry by CVE, GHSA, or OSV id, as JSON. Includes CISA's SSVC decision points when CISA published them. | `id` |
| `feed.read` | Lists the newest items in a feed. | `name` for one feed (`critical`, `all`, `kev`, `tech/<slug>`, or `ecosystem/<ecosystem>`), or any of `type`, `kind`, `filter`, `tech`, `ecosystem`, `vendor`, `fix` and `confidence`; `limit` sets how many items to return, from 1 to 200 |
| `stack.check` | Checks exact package versions against the catalog. | `lockfile_content` (the file's contents, with an optional `lockfile_name`), `packages` (a list of `ecosystem`, `name`, and `version`), or `stack_id` |
| `entities.search` | Searches vulnerabilities, early warnings, breaches, vendors, packages and technologies; an exact name match comes first. | `q`, with the same fields as the [search page](https://docs.sazabi.com/cve-now/api.md) |
| `<type>.get`, `<type>.list` | One entity by id, or a page of them, for `cve`, `early-warning`, `breach`, `vendor`, `package` and `tech`. | `id`, or the filters of the type with `limit` and `cursor` |
| `leaderboard.get`, `feedback.send` | The ranking of vendors, and a private note to the cve.now team. | see `search` |

An argument an operation does not take answers with an error that lists the ones it does.

The old tools `get_vulnerability`, `list_feed`, `check_stack`, `search` (with `q`), `get_leaderboard`, `get_package` and `send_feedback` are deprecated since 2026-11-11. They still answer as before, but `tools/list` no longer shows them, and each answer names the operation that replaces it.

`feed.read`'s `filter`, `tech`, and `ecosystem` work like the [feed parameters](https://docs.sazabi.com/cve-now/subscribe-to-feeds.md#build-a-feed-url): each takes one value or a list, an item must match any listed value of a field and every field passed, and at most 16 combinations are allowed.

`stack.check` accepts the same [lockfiles](https://docs.sazabi.com/cve-now/check-your-stack.md#supported-lockfiles) as an upload, and up to 10,000 packages per call. For a large lockfile, have the agent upload it and pass the returned id as `stack_id`, so the file's contents do not go through the agent's context:

```bash theme={"system"}
curl --data-binary @package-lock.json -H 'X-Lockfile-Name: package-lock.json' https://cve.now/api/v1/stacks
```

As with any stack check, no matches means cve.now found none, not that the packages are safe.

## Example prompts

* "Use cve.now to check this project's lockfile for vulnerabilities, and propose upgrades for anything critical."
* "What's on the cve.now critical feed today that affects Next.js or React?"
* "Look up CVE-2021-44228 on cve.now. Is it in KEV, and what version fixes it?"

## The Sazabi agent

The Sazabi agent has cve.now built in: it is a managed MCP connector that is on by default in every Sazabi project, with nothing to configure. Setup details are on the [cve.now connector page](https://docs.sazabi.com/catalogs/mcp-connectors/cve-now.md).

## Verify

Ask the agent to look up CVE-2021-44228 on cve.now. Confirm it runs `cve.get` and answers with the Log4Shell entry, including that it is in CISA KEV.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.