Prime Security Tools

Agent Skills, Claude Code plugins, and an MCP server for working with Prime Security from AI agents: security design reviews, policy search, repository analysis, code reviews, security posture, and AI-assisted security conversations.

What's Included

Component	What it does	Clients
prime skill	Calls the Prime API using its published route documentation	Claude Code, Codex, OpenCode, Cursor
Prime MCP server	Exposes Prime as MCP tools	Claude (web, Desktop, mobile), Claude Code, any MCP client
prime-security-operator plugin	The prime skill and the MCP server connection	Claude Code, Codex, Cursor
prime-security-dev	Portable code guardrails skill and optional client hooks; Claude plugin retains MCP connection	Claude Code, Codex, OpenCode, Cursor

Prerequisites

• A Prime Security account.
• HTTPS access to the Prime Security API.
• For the prime skill's browser login: python3 3.9 or later.
• For the Skills CLI installation only: Node.js 22.20.0 or later with npx, and HTTPS access to GitHub and npm.

On first use the prime skill opens your browser to log in to Prime Security, so the skill needs a Personal Access Token (PAT) only on machines without a browser. The development guardrails helper reads PRIME_PAT_TOKEN or the shared token file. MCP configurations except the Claude connector read PRIME_PAT_TOKEN; see Configure Credentials.

Install The Portable Skill

Use the Skills CLI for the primary cross-client installation:

npx skills add PrimeSecurityAI/prime-security-ai-plugins --skill prime --global

For a non-interactive installation targeting all four supported clients explicitly:

npx skills add PrimeSecurityAI/prime-security-ai-plugins --skill prime --global --agent claude-code --agent codex --agent opencode --agent cursor --yes

Choose one installation path. Do not install both the plain prime skill and the prime-security-operator Claude plugin for the same client; duplicate skills can produce ambiguous invocation and updates.

When Claude Code and OpenCode are targeted together, OpenCode may log a duplicate-source warning because it scans both ~/.agents/skills and Claude's compatibility directory. The Skills CLI links both paths to the same canonical installed copy, and OpenCode exposes one prime skill.

Update Or Remove

npx skills update prime --global --yes
npx skills remove prime --global --agent claude-code --agent codex --agent opencode --agent cursor --yes

Restart a client if it does not detect a newly installed, updated, or removed skill.

Configure Credentials

No setup is needed on a machine with a browser. When no credential is configured, the skill runs its browser login. After you log in and click Authorize, it saves a 30-day token to the token file below. The token appears under Settings > Access > API Token, where you can revoke it.

The skill resolves credentials in this order:

1. PRIME_PAT_TOKEN
2. ${XDG_CONFIG_HOME:-$HOME/.config}/prime-security/token

PRIME_API_URL is optional and defaults to https://api.primesec.ai.

On machines without a browser (SSH, CI, containers), create a PAT under Settings > Access > API Token > Create Token instead. Never paste a PAT into an agent chat. Paste this into Bash or zsh on macOS/Linux. Input is hidden. Empty or invalid input leaves the existing token unchanged:

bash -c '
set -e
LC_ALL=C
IFS= read -r -s -p "Prime PAT: " prime_pat || {
  [ "$?" -eq 1 ] && [ -n "$prime_pat" ]
} || {
  printf "\nNo token entered; existing token unchanged.\n" >&2
  exit 1
}
printf "\n"
prime_pat="${prime_pat%"$(printf "\r")"}"
prime_pat="${prime_pat#"${prime_pat%%[![:blank:]]*}"}"
prime_pat="${prime_pat%"${prime_pat##*[![:blank:]]}"}"
if [[ -z "$prime_pat" || "$prime_pat" == *[![:graph:]]* ]]; then
  printf "Invalid token; existing token unchanged.\n" >&2
  exit 1
fi
config_dir="${XDG_CONFIG_HOME:-$HOME/.config}/prime-security"
umask 077
mkdir -p "$config_dir"
chmod 700 "$config_dir"
token_tmp=$(mktemp "$config_dir/token.XXXXXX")
trap "rm -f -- \"\$token_tmp\"" EXIT
printf "%s\n" "$prime_pat" > "$token_tmp"
mv -f -- "$token_tmp" "$config_dir/token"
'

On other systems, use hidden terminal input and a current-user-only ACL for the same logical token path. Do not print or log the token.

The MCP configurations below read only the PRIME_PAT_TOKEN environment variable, not the token file. Export it in the shell profile or environment that starts your agent, for example from the token file:

export PRIME_PAT_TOKEN="$(cat "${XDG_CONFIG_HOME:-$HOME/.config}/prime-security/token")"

If it is unset, the server rejects the connection with 401.

MCP Server

Prime exposes a remote MCP server over Streamable HTTP at https://api.primesec.ai/mcp. Its tools act as the signed-in user, within that user's account and permissions.

Tools	Purpose
get_design_docs, get_design_doc_by_id	Security design reviews
get_repositories, get_repository_by_id, get_repository_code_issues	Code repositories and their findings
get_posture_metrics	Security posture metrics
get_all_products	Products
start_prime_ai_conversation, ask_prime_ai, get_prime_ai_answer	Ask Prime's security AI a free-form question, then poll get_prime_ai_answer until the answer is complete
search_prime_operations, call_prime_operation	Find and run any other Prime API operation

call_prime_operation can run operations that change data in your Prime account. Review those calls before approving them.

Claude (Web, Desktop, Mobile)

1. Go to Customize > Connectors, select + Add, then Add custom connector.
2. Enter https://api.primesec.ai/mcp as the URL.
3. In the OAuth client settings, use your own OAuth client with Client ID claude and no client secret.
4. Connect and sign in with your Prime account.

On Team and Enterprise plans, an organization Owner adds the connector, and each member connects and signs in. The connector is then available in Claude on the web, Desktop, and mobile, and in Claude Code.

Claude Code

Installing either Claude Code plugin connects the server, using PRIME_PAT_TOKEN and PRIME_API_URL when set. Installing both plugins still connects to it once.

Without a plugin, add it at user scope. This stores the token in your Claude Code configuration:

claude mcp add --transport http --scope user prime https://api.primesec.ai/mcp --header "Authorization: Bearer $PRIME_PAT_TOKEN"

Codex

The operator Codex plugin connects the server; see Codex Plugins. Without a plugin:

codex mcp add prime --url https://api.primesec.ai/mcp --bearer-token-env-var PRIME_PAT_TOKEN

Or add it to ~/.codex/config.toml:

[mcp_servers.prime]
url = "https://api.primesec.ai/mcp"
bearer_token_env_var = "PRIME_PAT_TOKEN"

OpenCode

Add it to ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "prime": {
      "type": "remote",
      "url": "https://api.primesec.ai/mcp",
      "oauth": false,
      "headers": {
        "Authorization": "Bearer {env:PRIME_PAT_TOKEN}"
      }
    }
  }
}

Cursor

Add it to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "prime": {
      "url": "https://api.primesec.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${env:PRIME_PAT_TOKEN}"
      }
    }
  }
}

Other MCP Clients

Connect to https://api.primesec.ai/mcp over Streamable HTTP and send Authorization: Bearer <PAT>. Keep the PAT in the client's secret or environment configuration, never in a chat.

Skill Invocation

All clients can select prime automatically from a natural-language request.

Client	Explicit invocation	Example
Claude Code	/prime	/prime review this design for security risks
Codex	$prime	$prime search our policies for password requirements
OpenCode	Ask it to use prime	Use the prime skill to review this design
Cursor	Ask it to use prime	Use the prime skill to analyze this repository

Manual Install Alternative

If the Skills CLI is unavailable, copy the canonical directory at plugins/prime-security-operator/skills/prime into the skill directory as a folder named prime:

Client	User skill directory
Claude Code	~/.claude/skills/prime
Codex	~/.agents/skills/prime
OpenCode	~/.agents/skills/prime
Cursor	~/.agents/skills/prime

For several clients, keep one copy at ~/.agents/skills/prime and link ~/.claude/skills/prime to it for Claude Code. On systems without links, use the client-specific directories. Do not combine manual installation with the Skills CLI or marketplace installation.

Claude Marketplace Alternative

The Claude marketplace remains available for backward compatibility, and its plugins also connect the MCP server:

claude plugin marketplace add PrimeSecurityAI/prime-security-ai-plugins
claude plugin install prime-security-operator@prime

Inside Claude Code, the equivalents are /plugin marketplace add PrimeSecurityAI/prime-security-ai-plugins and /plugin install prime-security-operator@prime.

The plugin skill is invoked as /prime-security-operator:prime in Claude Code.

Cursor Plugins

Both plugins also ship Cursor manifests, and .cursor-plugin/marketplace.json lists them. Each bundles its skill and the Prime MCP server, which reads PRIME_PAT_TOKEN from the environment that starts Cursor. Team admins can import this repository as a team marketplace. For a local install, copy the plugin directory (for example plugins/prime-security-operator) to ~/.cursor/plugins/local/<plugin-name>, then run Developer: Reload Window. Install a plugin or the Skills CLI skill for Cursor, not both. For development guardrails across Claude Code, Codex, OpenCode, and Cursor, see the prime-security-dev installation guide.

Codex Plugins

The Codex marketplace (.agents/plugins/marketplace.json) lists both plugins:

codex plugin marketplace add PrimeSecurityAI/prime-security-ai-plugins
codex plugin add prime-security-operator@prime

The operator plugin bundles the prime skill and the Prime MCP server, which reads PRIME_PAT_TOKEN and always uses https://api.primesec.ai/mcp; Codex does not expand PRIME_API_URL in plugin MCP URLs. Install a plugin or the Skills CLI skill for Codex, not both.

Plugins

prime-security-operator

The cross-client Prime API skill packaged for Claude Code, Codex, and Cursor, plus the MCP server connection. Its only canonical skill source is plugins/prime-security-operator/skills/prime/SKILL.md.

prime-security-dev

Portable prime-code-guardrails skill for Claude Code, Codex, and OpenCode, with safe shared credentials, a bundled API helper, and optional client lifecycle integrations. The Claude plugin retains its MCP server connection. Install separately from the operator skill; see the installation and support guide.

Structure

.claude-plugin/marketplace.json      Claude marketplace
.agents/plugins/marketplace.json     Codex marketplace
.cursor-plugin/marketplace.json      Cursor marketplace
plugins/
|-- prime-security-operator/
|   |-- .claude-plugin/ .codex-plugin/ .cursor-plugin/
|   |-- .mcp.json
|   `-- skills/prime/SKILL.md
`-- prime-security-dev/
    |-- .claude-plugin/ .codex-plugin/ .cursor-plugin/
    |-- .mcp.json
    |-- hooks/hooks.json codex.json cursor.json
    `-- skills/prime-code-guardrails/

License

MIT