# Security MCP server | Precursor Intelligence

**Ask your AI assistant about your own security. It answers with real data.**

> Our MCP server lets AI assistants such as Claude and Cursor use Precursor Intelligence as a set of tools. Ask a question and the assistant looks up your vulnerabilities, hosts, alerts and domains, plus current threat intelligence, then answers with the evidence. It is for IT and security teams who already work with an assistant.

Canonical page: https://www.precursorintelligence.com/features/mcp-server
Breadcrumb: [Home](https://www.precursorintelligence.com/) > [Features](https://www.precursorintelligence.com/features) > MCP server
Start free: https://www.precursorintelligence.com/register · Book a demo: https://www.precursorintelligence.com/contact

Server URL: `https://mcp.precursorintelligence.com` (Streamable HTTP, bearer API key). Works with Claude and Cursor. 71 tools. Scoped to your organisation.

Example data on this page uses the fictional company Acme (acme.co.uk, documentation-range IP addresses such as 203.0.113.24).

## Contents

1. [Which tools can your assistant use?](#which-tools-can-your-assistant-use)
2. [What happens when you ask?](#what-happens-when-you-ask)
3. [How do you connect it?](#how-do-you-connect-it)
4. [What can your assistant see?](#what-can-your-assistant-see)
5. [What could you ask it?](#what-could-you-ask-it)
6. [Questions](#questions)
7. [Related features](#related-features)

## Which tools can your assistant use?

MCP, the Model Context Protocol, is an open standard that lets AI assistants call outside tools. Ours has 71: 28 that read your own organisation's data (scope `mcp:org`, which you switch on per key) and 43 for public intelligence (scope `mcp:read`, on for every new key). Your assistant picks the right ones for each question.

### Your organisation (28 tools, `mcp:org`)

**Your attack surface**

- `list_my_assets`: your attack surface scans, with counts of hosts, services and flaws.
- `list_my_hosts`: hosts your scans found, with their IP addresses and open-port counts.
- `list_my_ports`: ports across your hosts, filtered by state, number or host.
- `get_open_ports`: the open ports, and the service on each, for one host or IP address.
- `list_my_services`: services answering on your hosts, with product and version.
- `list_my_ip_addresses`: IP addresses your scans found, with who runs the network.
- `list_my_network_ranges`: the network ranges linked to your organisation.
- `list_my_dns_records`: DNS records found for your domains, by record type.
- `list_my_discovered_domains`: domains your scans found, with registrar and expiry dates.
- `list_my_ssl_certificates`: certificates on your services, soonest to expire first.
- `list_my_software`: software found on your assets and your website.

**Your flaws and alerts**

- `assess_cve_exposure`: whether one CVE affects you, across your scans, tracked software and alerts.
- `list_my_vulnerabilities`: flaws your scans found, filtered by severity.
- `list_my_relevant_cves`: flaws matched to the software you track.
- `list_my_triage_findings`: your scored flaws with an action priority, highest first.
- `list_my_alerts`: your vulnerability alerts, newest first.
- `get_my_alert_config`: what you track and who receives your alert emails.
- `list_my_action_plans`: your AI action plans.
- `list_my_reports`: threat reports you have mapped to attack techniques.

**Your domains and people**

- `list_my_domains`: the domains you monitor for lookalikes.
- `list_my_domain_alerts`: alerts for new lookalike domains and changes to them.
- `list_my_exposed_credentials`: staff logins found in other companies' breaches. Passwords are never returned.

**Your code**

- `list_my_lockfiles`: the dependency files you have uploaded, with package counts.
- `check_my_supply_chain_exposure`: whether a hijacked or malicious package is in those files.

**Threats to you**

- `get_my_industry`: the industry set for your organisation.
- `list_my_favorite_threat_groups`: the threat groups you follow.
- `map_assets_to_actor`: which of your assets match a threat group's methods and flaws.
- `appliances_exploited_by_actor`: which of your firewalls, VPNs and gateways have flaws being exploited now.

### Public intelligence (43 tools, `mcp:read`)

**Vulnerabilities**

- `get_cve`: everything on one CVE, including the groups reported to exploit it.
- `search_cves_by_vendor`: a vendor's flaws, most likely to be exploited first.
- `list_recent_notable_events`: recent notable changes to flaws, such as a new exploited listing.
- `get_top_affected_systems`: the systems most written about in security reporting.
- `list_remediation_summary`: AI-written fix summaries for a CVE.
- `list_recent_remediation_queue`: flaws recently queued for a fix summary.

**Products and weaknesses**

- `get_vendor`: a vendor's products and how many flaws each has.
- `get_vendor_product`: a product's versions and the flaws in each.
- `get_vendor_product_version`: the flaws in one exact version.
- `search_products`: find a product by name.
- `list_top_vendors_by_cve_count`: the vendors with the most published flaws.
- `get_cwe`: one weakness type, with its consequences and fixes.
- `list_cwes`: search the catalogue of weakness types.
- `list_top_cwes`: the weakness types behind the most flaws.

**Threat groups**

- `list_threat_actors`: criminal and state-backed groups, by type.
- `get_threat_actor`: a group's full profile, each claim with a quote and source.
- `list_threat_actors_by_industry`: the groups reported to target an industry.
- `list_threat_actor_cves`: the flaws a group is reported to exploit.
- `get_cve_threat_actors`: the groups reported to exploit one CVE.
- `get_threat_actor_techniques`: how a group operates, technique by technique.

**Indicators and malware**

- `search_iocs`: search malicious IPs, domains, URLs and file hashes.
- `lookup_ioc`: check one exact IP, domain, URL or hash.
- `list_recent_iocs`: indicators first seen in the last few days.
- `search_malware_samples`: search malware samples by name, type or signature.
- `get_malware_sample`: one sample's details, by its file hash.
- `list_recent_malware_samples`: recently seen malware samples.
- `list_top_malware`: the malware families most written about.
- `lookup_ssl_blacklist`: whether a certificate fingerprint is linked to malware.
- `lookup_ja3_blacklist`: whether a connection fingerprint is linked to malware.

**Attack simulation tests**

- `list_art_techniques`: the techniques that have ready-made tests.
- `list_art_atomic_tests`: tests for a technique and platform.
- `get_art_atomic_test`: one test in full, with how to run it.

**Your account**

- `account_status`: your key details, rate limit and account balance.

**And ten more:** tools for the government list of exploited flaws, exploit-likelihood scores, attack techniques and what internet attack sensors are seeing. Their exact names are in the developer docs: https://docs.precursorintelligence.com

## What happens when you ask?

Your assistant does the talking. We supply the facts it looks up, and you only pay for the tool calls that return data.

1. **Ask a question (you).** Type it the way you would ask a colleague, such as "Are we exposed to CVE-2024-3400?"
2. **It picks the tools (your assistant).** It reads our tool list and calls the tools that fit the question, for example `assess_cve_exposure` and `get_cve`.
3. **We check the key and reply (Precursor Intelligence).** The key's scopes (`mcp:read`, `mcp:org`) decide which tools work, and only for your organisation.
4. **You get an answer with evidence (your assistant).** It answers from what the tools returned, such as the host affected and whether the flaw is being exploited.

Example (Acme): "Are we exposed to CVE-2024-3400?" The assistant calls `assess_cve_exposure` (confirmed, 1 host, vpn.acme.co.uk) and `get_cve` (VPN gateway flaw, known exploited), then answers: "Yes, one system. vpn.acme.co.uk runs the affected VPN gateway and the flaw is known to be exploited, so treat it as an emergency patch. Update it first, then check the device for signs of compromise."

## How do you connect it?

Add the server to your assistant once. It uses the same API key as our REST API, so there is nothing else to set up.

- Claude, Cursor or any MCP-capable assistant, connected over Streamable HTTP.
- Admins and Members create keys in Settings.
- Lock the key to your IP addresses and give it an expiry date. The secret is shown once and we keep only a hash.

Claude Code (run once in your terminal):

```
claude mcp add --transport http precursor-intelligence \
  https://mcp.precursorintelligence.com \
  --header "Authorization: Bearer <YOUR_API_KEY>"
```

Claude Desktop or Cursor (MCP server configuration):

```
{
  "mcpServers": {
    "precursor-intelligence": {
      "type": "http",
      "url": "https://mcp.precursorintelligence.com",
      "headers": { "Authorization": "Bearer <YOUR_API_KEY>" }
    }
  }
}
```

Direct HTTP (any MCP client, or a quick test):

```
curl -s https://mcp.precursorintelligence.com \
  -H "Authorization: Bearer $PRECURSOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"account_status","arguments":{}}}'
```

Setup steps for each assistant are in the developer docs: https://docs.precursorintelligence.com

## What can your assistant see?

Every key carries scopes. Public intelligence is on by default. Your own organisation's data is a separate switch, per key.

- `mcp:read` opens the 43 public intelligence tools. Every new key has it.
- `mcp:org` adds the 28 tools that read your scans, flaws, alerts, domains and code.
- The key decides the organisation. No request can ask for another organisation's data.
- Leaked passwords never reach the assistant. The exposed-logins tool says who is affected, not what their password was.
- The tools look things up. None of them start scans, unlock results or change your settings.

Every credential failure gets the same response, so nobody can probe for a working key.

## What could you ask it?

Questions in your own words, and the tools an assistant would reach for to answer them:

1. **Are we exposed to CVE-2024-3400?** Checks your scans, tracked software and alerts for that one flaw, then pulls its details. Tools: `assess_cve_exposure`, `get_cve`.
2. **What is open to the internet on 203.0.113.24?** Lists the open ports on that address and the service answering on each. Tool: `get_open_ports`.
3. **Which of our certificates expire soonest?** Lists the certificates on your services, soonest to expire first. Tool: `list_my_ssl_certificates`.
4. **Have any staff logins leaked?** Lists staff addresses found in other companies' breaches, without the passwords. Tool: `list_my_exposed_credentials`.
5. **Is anyone using a lookalike of our domain?** Shows the domains you monitor and the latest lookalike alerts for them. Tools: `list_my_domains`, `list_my_domain_alerts`.
6. **Is a hijacked package in our code?** Checks your uploaded dependency files against known supply-chain incidents. Tool: `check_my_supply_chain_exposure`.
7. **Who targets our industry, and do we run what they exploit?** Finds your industry, the groups reported to target it, and which of your assets match their methods and flaws. Tools: `get_my_industry`, `list_threat_actors_by_industry`, `map_assets_to_actor`.
8. **Is any firewall or VPN of ours running an exploited flaw?** Lists your edge devices tied to flaws that are being exploited, optionally for one named group. Tool: `appliances_exploited_by_actor`.

## Questions

### Which AI assistants does it work with?

Claude, Cursor and any other assistant that supports MCP over Streamable HTTP. You add one server URL, https://mcp.precursorintelligence.com, and your API key as a bearer token.

### Can the assistant see my organisation's data?

Only if its key has the mcp:org scope, which is off for new keys. Without it, the assistant can use the 43 public intelligence tools only. The key decides which organisation it reads, so it can never reach anyone else's data.

### Can it change anything in my account?

No. The tools look information up and return it to your assistant. None of them start a scan, unlock results or change settings.

### Is there a rate limit?

60 requests a minute per key, the same for every key.

## Related features

- [Integrations](https://www.precursorintelligence.com/features/integrations): a REST API, webhooks, Tenable sync and scan-file upload to connect the tools you already use.
- [Threat intelligence](https://www.precursorintelligence.com/features/threat-intelligence): who targets businesses like yours, how they work and which flaws they use.
- [Attack surface management](https://www.precursorintelligence.com/features/attack-surface-management): see what your business has open to the internet and what changed since last time.
- [Vulnerability management](https://www.precursorintelligence.com/features/vulnerability-management): tell us the software you run and see only the known flaws that affect it.
- [All features](https://www.precursorintelligence.com/features)

## Get started

Give your assistant real answers about your own estate. Add one URL to Claude, Cursor or any MCP assistant and ask about your vulnerabilities, hosts, alerts and domains, plus who is targeting your sector, and get answers backed by evidence.

- Start free: https://www.precursorintelligence.com/register
- Book a demo: https://www.precursorintelligence.com/contact
- Nothing to install · Unlimited team members · Data stored in London · Two-factor sign-in for everyone