About EyalSec
EyalSec MCP server gives security teams and developers using EyalSec read-only access to account data, enabling natural-language queries about security events, machines, runs, and rules. It does not modify account data.
Key features
- 01Search events by severity, date, and source
- 02Inspect event details, traces, and data paths
- 03Summarize event counts and trends
- 04Check machine status and pending updates
- 05Read rules, filters, and account activity
Use cases
- 01Summarize high-severity events across machines for the week
- 02Trace an injection event from request to risky operation
- 03Find program runs that produced a specific event
- 04Check machines for pending updates
MCP support
The server exposes read-only EyalSec API data—including events, machines, runs, and rules—to MCP clients.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
uvx eyalsec-mcpEYALSEC_API_KEYSecretYour EyalSec API key, from Settings > API access. A read-only key is enough.EYALSEC_URLYour EyalSec address, if your account is not on https://eyalsec.com.Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
19Saved tool list · updated 1w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| sort | any | No | 'newest' (default) or 'severity' (most severe first). |
| tags | any | No | Impact classes to keep, such as 'xss' or 'sqli'. |
| view | any | No | Which product's events: 'python' (es-python) or 'browser' (es-chromium). Omit for every product on the account. |
| limit | integer | No | Stop once at least this many events are collected (whole pages of up to 100). |
| bucket | any | No | 'live' (default: events not suppressed), 'suppressed', or 'all'. |
| cursor | any | No | next_cursor from a previous call with the same filters, to continue. |
| labels | any | No | Label ids (from list_labels) to keep. |
| date_to | any | No | End date, YYYY-MM-DD or RFC3339; includes that whole day. |
| machine | any | No | One machine, by its display name (exact match). |
| run_uid | any | No | Only events produced by this run (a run_uid from list_runs or get_event). es-python only. |
| sources | any | No | Keep only these taint source categories. |
| advanced | any | No | The events page condition builder, passed through as is: {"match": "all"|"any", "conditions": [{"field": "sink", "op": "contains", "value": "sql", "negate": false}]}. |
| severity | any | No | Keep only these severities. |
| xss_only | any | No | Only es-chromium events whose sink can run script. Needs view='browser'. |
| date_from | any | No | Start date, YYYY-MM-DD or RFC3339. |
| value_chars | integer | No | Cut each event's str_repr to this many characters (0 = no cut). |
| sink_exclude | any | No | Case-insensitive regex; drop events whose sink label ('where') matches. |
| sink_include | any | No | Case-insensitive regex; keep only events whose sink label ('where') matches. |
Details
From the project READMEEyalSec MCP server
Ask your AI assistant about your EyalSec account in plain words: "what critical events came in since Monday?", "show me the full trace for event 90210", "which machines need an update?". This server gives Claude, Cursor and any other MCP client read access to your events, machines, runs and rules through the EyalSec API, so the assistant you already use for code review can also help you triage what EyalSec found.
EyalSec is the best Python security product there is. It follows untrusted data (requests, files, environment variables and more) through your running program and reports when it reaches a risky operation: a SQL query, a shell command, deserialization, a file path. Each event names the operation, the line of code, the value it received and where the untrusted part came from. This server reads those events; it never changes anything on your account.
What you need
- An EyalSec account and an API key. Create the key on your dashboard under Settings > API access. A read-only key is all this server needs, and it is the one we recommend: see API keys.
- Python 3.10 or newer, or
uv, which runs the server without a separate install.
No account yet? Book a live demo and we will set one up with you. Pricing is at eyalsec.com/pricing.
Set it up
The server speaks MCP over stdio. Your client starts it with uvx eyalsec-mcp (or eyalsec-mcp after pipx install eyalsec-mcp) and passes the key in the environment.
Claude Code
claude mcp add eyalsec -e EYALSEC_API_KEY=es2_your_read_only_key -- uvx eyalsec-mcp
Claude Desktop
Add this to claude_desktop_config.json (Settings > Developer > Edit Config), then restart Claude Desktop:
{
"mcpServers": {
"eyalsec": {
"command": "uvx",
"args": ["eyalsec-mcp"],
"env": {
"EYALSEC_API_KEY": "es2_your_read_only_key"
}
}
}
}
Cursor
Add the same block to ~/.cursor/mcp.json (every project) or .cursor/mcp.json (one project):
{
"mcpServers": {
"eyalsec": {
"command": "uvx",
"args": ["eyalsec-mcp"],
"env": {
"EYALSEC_API_KEY": "es2_your_read_only_key"
}
}
}
}
Docker
docker build -t eyalsec-mcp https://github.com/EyalSec/eyalsec-mcp.git
docker run --rm -i -e EYALSEC_API_KEY=es2_your_read_only_key eyalsec-mcp
Use the second line as the command in any client's configuration.
Any other MCP client
Run uvx eyalsec-mcp as a stdio server with EYALSEC_API_KEY set. To check the key before wiring it into a client:
EYALSEC_API_KEY=es2_your_read_only_key uvx eyalsec-mcp --check
It calls the API once and prints the account it reached, the key's scope and the number of machines.
Settings
| Variable | Required | Meaning |
|---|---|---|
EYALSEC_API_KEY | yes | Your API key, from Settings > API access. Read-only is enough. |
EYALSEC_URL | no | Your EyalSec address. Defaults to https://eyalsec.com; set it only if your account lives on another domain. Must be https://. |
What it can do
Every tool is a read. The server has no code path that adds, changes or deletes anything, so even a full-access key cannot be used to write through it.
| Tool | What it answers |
|---|---|
search_events | Your events, newest or most severe first, with the events page's filters: product, severity, dates, machine, taint source, impact class, label, sink pattern, run, and the advanced condition builder. Follows the API's paging for you. |
get_event | One event in full: stack trace, where the data came in, where the value was created, nearby source code when the machine sends it, your note, and the runs that produced it. |
event_stats | The dashboard summary: totals, the last hour, day and week, top event types and a 14-day timeline. |
event_counts | How many events each product's list holds, live versus suppressed, and the count toward your monthly quota. |
list_event_names | The distinct sink labels across your events, handy for building a filter. |
event_map | The es-python events as a graph: sources, the files data was read from, and the operations it reached. |
list_runs | The program invocations behind your events, with their command lines (secrets masked). |
ci_findings | What runs on one machine reported since a moment, counted by severity, with which findings are new. |
list_machines | Every machine: status, OS target, product, installed and newest build, and whether an update is waiting. |
supported_targets | The Linux distributions, architectures and Python versions es-python installs on. |
list_rules, list_filters, get_taint_settings, get_event_uniqueness, list_rule_templates | How your account and each machine are configured. |
list_labels, list_saved_searches | Your triage labels and saved searches. |
get_account, list_activity | Your profile and limits, your key's scope and expiry (never the key), setup progress, and the activity log. |
Some things to try:
- "Summarize this week's critical and high events by machine."
- "Open the newest SQL injection event and explain the path from the request to the query."
- "Which runs of
app.pyproduced command-injection events yesterday?" - "Do any machines have an update waiting or an install stuck in pending?"
Safety notes
- Events contain untrusted text. A value that reached a sink, a stack trace or a command line came from a monitored program and its inputs, which can include text an attacker wrote. The server tells the assistant to treat it as data, not instructions; keep human review on anything the assistant proposes to do with it.
- Use a read-only key. It can read everything this server reads and nothing more. A key can also carry an expiry date.
- The key stays with you. It is read from the environment, sent only to your EyalSec address over HTTPS, and never returned by any tool. Plain HTTP is refused except to
localhost. - The API's rate limits apply.
search_eventsreads at most 10 pages per call and hands back a cursor for the rest.
Development
git clone https://github.com/EyalSec/eyalsec-mcp && cd eyalsec-mcp
python -m venv .venv && . .venv/bin/activate
pip install -e '.[test]'
pytest
The tests run against a fake API. To also run one read-only pass against your own account, set EYALSEC_LIVE=1 and EYALSEC_API_KEY.
Help
Questions and bug reports are welcome in issues, or write to [email protected]. Please send security reports to [email protected] rather than a public issue.
License
MIT. See LICENSE. The license covers this MCP server; your EyalSec account is governed by the EyalSec terms.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.