Skip to main content

MCP Server

The FireMon MCP server exposes your FireMon environment to your own AI system. Connect any client that speaks MCP over Streamable HTTP, and your model can query firewall rules, devices, compliance data, metrics and Policy Planner workflows directly, and write and run its own SIQL.

Your model does the reasoning and writes the queries. FireMon serves the data and enforces your existing permissions. Nothing you send over MCP goes to a FireMon model, with one exception: search_documentation embeds your question with FireMon's OpenAI account to search the public documentation index. The other tools call no model at all.

The server works whether Insights AI is turned on or off for your organization. That setting governs FireMon's own AI features. Over MCP, your model does the work.

There are two levels of access:

  • Without credentials you get one tool, search_documentation, which searches FireMon product documentation. Nothing to set up.
  • With credentials you get the full tool set against your own environment: firewall rules, devices, device groups, compliance controls, Policy Planner workflows, metrics, and arbitrary SIQL queries.

Endpoint​

https://mcp.prod.firemon.cloud/mcp

POST only. Clients configured with the earlier address, https://4c5y9c4cyg.execute-api.us-west-2.amazonaws.com/mcp, keep working: it forwards to the same server. Requests through it are cut off after 29 seconds, though, so move to the address above.

Documentation search, no credentials​

Any MCP client can reach search_documentation without authenticating. It searches FireMon product documentation and returns the closest matching passages with their source URLs.

In Claude Code:

claude mcp add --transport http firemon-docs https://mcp.prod.firemon.cloud/mcp

With curl:

curl -s -X POST 'https://mcp.prod.firemon.cloud/mcp' \
-H 'content-type: application/json' \
-H 'accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_documentation","arguments":{"question":"How do I add a device group?"}}}'

The accept header must list both content types. Leave one out and the request is rejected with 406 before it reaches any tool.

Connecting to your own data​

You need two things: an Insights API key, and the ID of a Security Manager user for your agent to act as.

The API key​

The key is an administrator credential. It stands for your whole organization, and whoever holds it can act as any of your synced users, so store it the way you would store an administrator password. API Keys covers what a key authorizes and how to review and revoke one.

Don't try to reuse the key your SIP installation holds. SIP receives a fresh one every 24 hours, and Insights never shows it. Issue a long-lived key of your own instead:

  1. Sign in to Insights as a user with permission to view and modify client API keys, and open https://insights.firemon.cloud/setup. Nothing in the navigation links to it, so type the path.

  2. Click Generate Code. A short code appears. It can be used once, and expires after about a day.

  3. Exchange the code for the key:

    curl -s -X POST 'https://insights.firemon.cloud/api/rpc/register/exchangeCodeForApiKey' \
    -H 'content-type: application/json' \
    -d '{"code":"YOUR_CODE"}'

    The response carries the key:

    {"result":{"data":{"apiKey":"..."}}}
  4. Store the key now. Insights keeps only a hash of it, so this response is the only time it is visible. If you lose it, generate a new code and issue another key.

The new key appears in the list on the Setup page, identified by its ID. That list is also where you revoke it.

The SIP user ID​

Every call runs in Security Manager as a user you name in the x-sip-user-id header, and Security Manager applies that user's permissions as it would in its own UI. The value is the numeric ID Security Manager assigns a user account, not a username, an email address or an Insights login.

Create a dedicated account for the agent in Administration > Access > Users, and put it in a user group that grants only the device groups and permissions the agent should have. Everything the agent does is then attributed to that account in Security Manager, and you can disable it without touching anyone else. The agent never signs in anywhere, so the account needs no Insights access of its own.

The account has to be one SIP has synced to Insights. SIP sends its users across on its own schedule, so an account created moments ago is rejected until the next sync, usually a matter of minutes. Users and Access explains the sync. An account with no authorized device groups is refused as well.

To find the ID, list your users through the Security Manager API, which returns each user with its id:

GET https://YOUR_SIP/securitymanager/api/domain/1/user?pageSize=100

Replace 1 with your domain ID if you use more than one domain.

Configure your client​

Point your MCP client at the endpoint over Streamable HTTP and set these headers on every request:

HeaderValue
x-api-keyYour Insights API key
x-sip-user-idThe ID of the Security Manager user the agent acts as
x-device-group-idOptional. The device group that scopes calls which take no explicit one. It must be a group that user is authorized for. Without it, the user's first authorized group is used.
acceptapplication/json, text/event-stream — both types, or the request is rejected

In Claude Code:

claude mcp add --transport http firemon https://mcp.prod.firemon.cloud/mcp \
-H 'x-api-key: YOUR_API_KEY' \
-H 'x-sip-user-id: YOUR_SIP_USER_ID'

The server is stateless: no sessions, no resumability, and responses are always JSON rather than an event stream. Every request stands on its own.

Confirm it works​

List the tools you have access to:

curl -s -X POST 'https://mcp.prod.firemon.cloud/mcp' \
-H 'content-type: application/json' \
-H 'accept: application/json, text/event-stream' \
-H 'x-api-key: YOUR_API_KEY' \
-H 'x-sip-user-id: YOUR_SIP_USER_ID' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

A long list means you're authenticated. A single tool means your x-api-key header isn't reaching the server at all. A key that arrives but isn't recognized returns 401 instead, rather than falling back to documentation-only access.

tools/list is also the authoritative reference. It returns the current description and full JSON Schema for every tool, which will always be more current than this page.

Start with your device groups​

Most tools are scoped to a device group, and the IDs aren't guessable, so list them first:

curl -s -X POST 'https://mcp.prod.firemon.cloud/mcp' \
-H 'content-type: application/json' \
-H 'accept: application/json, text/event-stream' \
-H 'x-api-key: YOUR_API_KEY' \
-H 'x-sip-user-id: YOUR_SIP_USER_ID' \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"list_all_device_groups","arguments":{}}}'

You'll get back each group's id, name, and device counts. Device group 1 is "all devices".

From there you have two ways to set the scope:

  • Pass deviceGroupId in the arguments of tools that accept it.
  • Send an x-device-group-id header to set the default for every call. If you don't, the default is the first group your user is authorized for.

Either way the group has to be one your SIP user is authorized for, or the call is refused.

The tools​

Arguments marked optional can be left out or sent as null; both mean "not set". Everything else is required.

Firewall rules​

ToolWhat it doesArguments
count_firewall_rulesCounts rules matching your criteria. Use it for counting and existence checks.sourceIp, destinationIp, portNumberOrServiceName, actionType (ALLOW/DENY), includeWildcardRules, deviceId, deviceName, deviceGroupId, deviceGroupName — all optional
search_firewall_rules_by_sourceFinds rules by source address only.sourceIp; portNumberOrServiceName, actionType optional
search_firewall_rules_by_destinationFinds rules by destination address only.destinationIp; portNumberOrServiceName, actionType optional
list_all_security_rules_by_deviceReturns every rule on a device.deviceId optional — leave it out for every device

For source and destination, the literal string PUBLIC_INTERNET matches traffic from or to the internet.

Devices and device groups​

ToolWhat it doesArguments
list_all_device_groupsLists every device group with IDs, names, and counts.none
list_all_devicesLists devices in a group.deviceGroupId optional — defaults to your scope
get_device_group_kpiKPIs for a group: control failures, recently changed devices, rule usage.kpi, id
get_device_kpiKPIs for one device: control failures, security concern index, config changes, complexity.kpi, deviceId

get_device_group_kpi accepts CONTROL_FAILURES_WITH_HIGHEST_SEVERITY, DEVICES_MOST_RECENTLY_CHANGED, or TOTAL_RULE_USAGE_CHART.

get_device_kpi accepts CONTROL_FAILURES_WITH_HIGHEST_SEVERITY, KPI_AVERAGE_SCI, KPI_TOTAL_CONFIG_CHANGES, or KPI_DEVICE_COMPLEXITY.

Network​

ToolWhat it doesArguments
get_network_nodeId_for_ip_addressFinds the network node an address or range sits behind, with its name.ipAddress

More than one node coming back means the network map couldn't be read cleanly for that address. Path simulation is not offered over MCP.

Metrics and compliance​

ToolWhat it doesArguments
query_metric_dataCurrent metrics for every device in a group, or a time series for one metric.queryType (deviceGroupMetrics or historicalMetrics); deviceGroupId, metricName, startDate, endDate, interval optional
get_metrics_summaryAggregated metrics across a date range, with start, end and percent change.startDate, endDate, interval; deviceGroupId optional
get_assessment_statsPer-control pass/fail counts over a range, with current, start, and prior-period values.startDate, endDate; deviceGroupId optional

Dates are epoch milliseconds. interval is day, week, month, or year. Metrics data can be up to 24 hours old; everything else on this page is real-time.

Every metrics query is scoped by deviceGroupId. Metrics for one specific device, and the per-workflow Policy Planner series, are not available over MCP; Policy Planner metrics are returned for the device group as a whole.

Policy Planner​

ToolWhat it doesArguments
list_policy_planner_workflowsLists workflows and their IDs.none
get_policy_planner_workflow_tasksReturns a workflow's stages and configuration, including auto-design settings.workflowId

To query tickets themselves, use execute_siql_query against /policyplanner/api/siql/ticket/paged-search.

SIQL​

SIQL is FireMon's query language. Between them, these two tools reach security rules, devices, device groups, network and service objects, policies, NAT rules, controls, assessments, assets, and Policy Planner tickets.

It's a two-step job: pull the reference for the section you need, then write and run the query. Your own model writes the query on your side. FireMon never sees the request, only the finished query.

SIQL queries aren't limited by the device group header. Neither are list_all_device_groups, list_all_security_rules_by_device, get_device_kpi, or the Policy Planner tools — these run as your SIP user, and Security Manager applies that user's own permissions instead.

ToolWhat it doesArguments
get_siql_referenceReturns SIQL reference documentation by section.sections (array)
execute_siql_queryRuns a SIQL query against a Security Manager or Policy Planner endpoint.query, endpoint; countType, page, pageSize optional

Sections available to get_siql_reference: overview, operators, functions, pipes, rule, natrule, device, devicegroup, domain, policy, networkobj, serviceobj, userobj, appobj, profileobj, control, assessment, connection, application, addressbook, scope, scheduleobj, interface, route, asset, ticket, examples. Start with overview if you're new to it.

endpoint must be one of:

/securitymanager/api/siql/addressbook/paged-search
/securitymanager/api/siql/application/paged-search
/securitymanager/api/siql/applicationgroup/paged-search
/securitymanager/api/siql/appobj/paged-search
/securitymanager/api/siql/assessment/paged-search
/securitymanager/api/siql/asset/paged-search
/securitymanager/api/siql/control/paged-search
/securitymanager/api/siql/device/paged-search
/securitymanager/api/siql/devicegroup/paged-search
/securitymanager/api/siql/interface/paged-search
/securitymanager/api/siql/natrule/paged-search
/securitymanager/api/siql/networkobj/paged-search
/securitymanager/api/siql/policy/paged-search
/securitymanager/api/siql/profileobj/paged-search
/securitymanager/api/siql/route/paged-search
/securitymanager/api/siql/scheduleobj/paged-search
/securitymanager/api/siql/scope/paged-search
/securitymanager/api/siql/secrule/paged-search
/securitymanager/api/siql/serviceobj/paged-search
/securitymanager/api/siql/urlmatcher/paged-search
/securitymanager/api/siql/userobj/paged-search
/securitymanager/api/siql/vendortag/paged-search
/securitymanager/api/siql/count
/policyplanner/api/siql/ticket/paged-search

countType is required when using /siql/count — it names the entity being counted: SECURITYRULE, NATRULE, DEVICE, DEVICEGROUP, NETWORKOBJECT, SERVICEOBJECT, USEROBJECT, APPLICATIONOBJECT, PROFILEOBJECT, CONTROL, ASSESSMENT, ASSET, POLICY, SCHEDULEOBJECT, INTERFACE, or ROUTE.

page is 0-indexed and defaults to 0. pageSize defaults to 20.

Running a query:

curl -s -X POST 'https://mcp.prod.firemon.cloud/mcp' \
-H 'content-type: application/json' \
-H 'accept: application/json, text/event-stream' \
-H 'x-api-key: YOUR_API_KEY' \
-H 'x-sip-user-id: YOUR_SIP_USER_ID' \
-d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"execute_siql_query","arguments":{"query":"device{active=true}","endpoint":"/securitymanager/api/siql/device/paged-search"}}}'

Product help​

ToolWhat it doesArguments
search_documentationSearches FireMon product documentation.question
get_sip_ui_navigationReturns the SIP site map with navigation paths, for answering "where do I find X in the UI?"none
list_available_shortcutsDescribes the hashtag shortcuts of the in-app Ask AI assistant. They are a feature of that chat window, so there is nothing for an MCP client to invoke.none

Argument validation​

Every call is checked against the tool's schema before anything runs. tools/list shows the exact schema the server enforces.

  • Optional means omit or null. An argument whose type allows null may be left out or sent as null, and the tool treats the two the same. An argument whose type does not allow null has to be present.
  • Types and enums are enforced. page is a number, so "page": "1" is rejected. kpi and endpoint must be one of the listed values.
  • Unknown argument names are refused, with Unrecognized key: "<name>". Four names on the metrics tools get a longer message, because the chatbot accepts them and MCP does not: deviceGroupName, workflowId, deviceId and deviceName on query_metric_data, and workflowId on get_metrics_summary. For those, the validation error's detail reads Access denied: <name> selects data the device-group check cannot cover … and names the alternative. Use list_all_device_groups to turn a group name into an ID, and scope the call with deviceGroupId.

A failed check comes back as a tool error, not an HTTP error. An invalid argument reads MCP error -32602: Input validation error: Invalid arguments for tool <name>: <detail>, and an unknown tool name reads MCP error -32602: Tool <name> not found.

Limits​

Each request has 50 seconds. A single slow query to Security Manager is given 45 seconds. Tools that page through large result sets — list_all_devices, list_all_device_groups, list_all_security_rules_by_device, list_policy_planner_workflows, get_assessment_stats — can run out of time on a large environment and return <tool> timed out.. If they do, narrow the request: a specific device group, a tighter SIQL query, a smaller pageSize. Retrying the same call unchanged will time out again.

Rate limit is 100 requests per second, shared across everyone using the server. One agent question can fan out into a dozen tool calls, so pace a busy agent.

Through the earlier execute-api address the limits are 29 seconds and 50 requests per second, bursting to 100, and a slow call surfaces as a gateway error rather than a tool error.

Troubleshooting​

What you seeWhat it meansWhat to do
Only search_documentation in tools/listYour x-api-key header isn't reaching the serverCheck the header name and that your client is actually sending it
401 Invalid x-api-keyThe key isn't recognized, or it expiredConfirm the key is current. If it worked yesterday and fails today, it was a 24-hour key — issue a long-lived one from the Setup page
403 Missing x-sip-user-id headerKey accepted, but no user namedAdd the x-sip-user-id header
403 Unknown SIP userThat ID hasn't been synced to InsightsCheck the ID in Security Manager, and give a new account a few minutes to sync
403 SIP user is disabledThe account is disabled in SIPRe-enable it in Administration > Access
403 SIP user has no authorized device groupsThe account has no device group accessPut it in a user group that grants at least one device group
403 Device group N is not authorized for this userYour x-device-group-id names a group the user can't seeUse list_all_device_groups to see what's available
406The accept header is missing a typeSend accept: application/json, text/event-stream
429Rate limitedBack off and retry
Input validation error: Invalid arguments for tool …An argument has the wrong type, an unlisted enum value, or an unknown nameCompare the call with the schema from tools/list
Access denied: deviceGroupName selects data …You passed an argument MCP withholdsScope the call with deviceGroupId instead
Access denied: device group N is not authorized for this userThe deviceGroupId in your arguments isn't one of yoursCheck the ID against list_all_device_groups
<tool> timed out.The call exceeded 50 secondsNarrow the request
A gateway error or a cut-off after 29 secondsYou're using the earlier execute-api addressSwitch to https://mcp.prod.firemon.cloud/mcp