AI integrations
MCP tool reference
Every tool on the hosted MCP server: what it answers, whether it reads or writes, and when it is listed. Generated from the tool catalog.
The hosted MCP server offers 41 tools. 38 are listed by default. The 3 write tools (propose_operations, submit_conversion, apply_mapping_answers) are listed only when NetConverter enables them for your tenant, and never in strict view. In strict view, 16 tools are listed. get_account reports how many tools your session has.
This page is generated from the tool catalog in the NetConverter code, so it changes when the catalog changes. For setup, sign-in and view modes, read the MCP server.
How job tools work
Most tools answer about one conversion job. Call list_jobs to see your recent jobs, then select_job with a job id; the selection is kept for 24 hours. You can also pass job_id to a single call. A job that does not exist, belongs to another tenant or has not been selected returns the same message.
Results follow your view mode: in the default view, IP addresses and CIDRs from your own jobs are shown; in strict view they are replaced with <ip> and <cidr>. Credential values are redacted in every view, and each result is limited to 64 KB.
Reading the tables
| Column | Meaning |
|---|---|
| Kind | Read: answers a question and changes nothing. Session setting: stores your selected job or session view level, and changes no job. Write: starts new work, such as a conversion. |
| Needs a job | Whether the tool reads one job. Tools marked No work without one. |
| Listed | Default and strict: listed in both view modes. Default view: hidden in strict view. When enabled: a write tool, listed only when NetConverter enables the write tools for your tenant, never in strict view. |
5 tools (kb_search, kb_explain, get_job_brief, search, fetch) send the query text, with secrets removed, to an outside embedding provider to search the knowledge base.
Knowledge base and search
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
kb_searchSearch the migration knowledge base | Searches the NetConverter knowledge base for cross-vendor patterns, mappings and guidance. Takes query and an optional top_k of 1 to 10. | No | Read | Default and strict |
vendor_capability_lookupVendor capability lookup | Whether a feature is supported on a vendor platform. Takes vendor, feature and optional technology. Vendors: cisco_asa, palo_alto_set, palo_alto_xml, palo_alto_panorama, fortinet, cisco_ios_xe, juniper_junos, aruba_cx, aruba_os, cisco_fmc. | No | Read | Default and strict |
kb_explainExplain from the knowledge base | Explains a config entity, selected lines or a feature using knowledge base documents, with citations and a coverage verdict of strong, weak or none. Takes entity and optional focus (explain, use_cases, best_practices, all). | No | Read | Default and strict |
searchSearch jobs and knowledge | Searches the knowledge base and your jobs. Returns up to 20 results as id, title, url. Ids are kb:<doc id> or job:<job id>. The url values are NetConverter identifiers, not web links. This tool and fetch follow the shape that ChatGPT connectors require. | No | Read | Default and strict |
fetchFetch a search result | Returns one search result by id. A kb: id returns a knowledge base summary of up to 4 KB. A job: id returns the governed job summary: status, verdict, counts, abstracted findings and the list of evidence sections. | No | Read | Default and strict |
Job selection
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
list_jobsList my jobs | Recent conversion jobs for this account: job id, vendors, device type, status, created time. Optional limit of 1 to 25, default 10. | No | Read | Default and strict |
select_jobSelect a job | Binds a job to your credential for 24 hours so later calls need no job_id. Returns the job's vendors and status. | No | Session setting | Default and strict |
set_view_levelLower this session's view | Lowers the view for this MCP session for 24 hours: none, masked or selected give strict view, full gives your grant's view. Never shows more than your grant allows, and cannot be raised again in the same session. See View modes. | No | Session setting | Default and strict |
Evidence and reports
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
get_target_configTranslated configuration summary | The post-translation target sections with counts and samples. Available only when a target result exists, so a blocked, validate-only or in-progress job returns a governed "not available" result. It returns samples, not the full configuration file. Hidden in strict view. | Yes | Read | Default view |
get_evidenceValidation evidence | The trust summary for a conversion. With no arguments: translation status, overall verdict and confidence, per-checkpoint pass or fail and severity counts, behavior check verdicts, NAT and flow proof counts, finding counts with abstracted titles, and an index of sections. Pass section to drill into one. section='unparsed_lines' lists the source lines the parser could not read, with their credential-redacted, address-tokenized text (at most 240 characters each); it is refused in strict view mode. Returns a governed "not available" result when evidence was not generated. | Yes | Read | Default and strict |
why_needs_reviewWhy this job needs review | One ranked answer to "why does this job need review?": blocking reasons for a blocked job, then each readiness check that holds the job (critical, review, advisory) with a count, up to five items, what to do and the get_evidence section to read next. basis is ledger for jobs converted on the current engine and legacy_text_match for older jobs, which are explained from their warning text. It never changes a job's status. | Yes | Read | Default and strict |
get_reportJob report | The validation or optimization report. report_type is validation (findings and warnings) or optimization (unused objects, redundant rules, cleanup). Returns a governed "not available" result when that report was not produced. | Yes | Read | Default and strict |
compare_against_intentCompare against intent | A deterministic comparison of the job against an intent that is attached to it. It refuses if no intent is attached. Optional focus. | Yes | Read | Default view |
get_job_briefJob overview | A job artifact by focus: evidence, optimization, target or best_practice. Optional query. | Yes | Read | Default and strict |
get_diagramGet a branded diagram | Source or target diagrams bound to stored job evidence. Returns availability, coverage, provenance and authenticated download descriptors for SVG, PNG, PDF, editable draw.io and JSON. The download checks the expected evidence hash and current access/view level. Missing evidence remains unavailable; configuration-derived relationships do not prove live topology or traffic. | Yes | Read | Default and strict |
Configuration summaries
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
get_interface_summaryInterface summary | Interface inventory: name, nameif, address, prefix, security level, zone, status, VLAN. | Yes | Read | Default view |
get_route_summaryRouting summary | Static routes (destination, gateway, administrative distance, metric), OSPF processes, BGP neighbors, EIGRP. | Yes | Read | Default view |
get_nat_summaryNAT summary | NAT rules with original and translated source and destination, services, zones, and an aggregation by type. Direction is given per rule. | Yes | Read | Default view |
get_acl_summaryAccess-list summary | Up to 200 rules with action, zones, addresses and service, with aggregations by action and zone. The total count is always reported. | Yes | Read | Default view |
get_zone_summarySecurity zone summary | All zones and their member interfaces. | Yes | Read | Default view |
get_object_summaryAddress and service object summary | Networks, network groups, services and service groups, up to 100 each with full totals. Pass object_type (networks, network_groups, services, service_groups, all) to get one type complete, up to 500. | Yes | Read | Default view |
get_vpn_summaryVPN summary | IPsec tunnels (peer, subnets, IKE version, crypto, lifetime, PFS) and IKE policies. data_quality is empty when there is no VPN. | Yes | Read | Default view |
get_management_configManagement settings summary | Hostname, domain, NTP, SNMP, syslog and AAA. Credentials such as SNMP communities are redacted, with a count kept. | Yes | Read | Default view |
get_unused_objectsUnused objects | Objects that no rule or group references, with a total and a percentage. Top 100 names per section. | Yes | Read | Default view |
get_object_dependenciesObject dependencies | Everything that references object_name, one hop, with a safe_to_delete flag. | Yes | Read | Default view |
query_irQuery the configuration model | A natural-language query over the parsed configuration, up to 50 matches plus aggregations. A fallback for when no purpose-built tool fits. Query up to 500 characters. | Yes | Read | Default view |
list_objects_with_attributeFind objects by attribute | Objects by attribute: ip_address, subnet, name_pattern, type, zone, interface or tag. Takes attribute and value. | Yes | Read | Default view |
list_rules_matchingFind matching rules | Security rules that match a predicate: src_zone, dst_zone, src_object, dst_object, service, action (allow or deny). | Yes | Read | Default view |
Flow simulation
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
simulate_flowSimulate a traffic flow | A deterministic packet trace through ingress, zone, route, NAT, layer 4, application and rule evaluation. Returns a verdict and an evidence chain. source_ip and destination_ip accept an address or an object or host name from the job. A group name returns needs_disambiguation, and you re-call with a member. Optional protocol (default tcp), destination_port, source_port and application. The application field feeds App-ID on Palo Alto and is ignored for ASA and Fortinet. Full support for ASA, Palo Alto and Fortinet. | Yes | Read | Default view |
Graph
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
get_dependency_chainsDependency chains | Transitive dependencies of object_name, depth 1 to 3, default 2. Prefer this over get_object_dependencies for relationship questions. | Yes | Read | Default view |
get_critical_object_impactCritical object impact | Blast radius of object_name: direct and indirect references, affected rule and NAT counts, and a severity of low, medium, high or critical. | Yes | Read | Default view |
get_translation_edge_summaryHow objects were translated | Lineage of object_name from source to target objects, with confidence and job id. limit 1 to 50, default 10, and an average confidence. | Yes | Read | Default view |
Guided mapping
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
get_mapping_questionsMapping questions for a job | The decisions for a completed conversion or prescan: the target zone name for each source zone, the target port for each source interface, and options such as zone naming, the Palo Alto routing engine, an explicit deny-all rule, logging on all rules and cleanup. Each question has a decision_id, its options, the value the job used, a recommended value and the reason. | Yes | Read | Default and strict |
validate_mappingCheck mapping answers | Checks answers [{decision_id, value}] without applying them and lists errors by decision_id (unknown name, an interface where a zone is required, two interfaces on one port, an invalid value). | Yes | Read | Default and strict |
apply_mapping_answersApply mapping answers and re-convert | Validates the answers and, when all are valid, starts a new conversion of the same source with them. Returns the new job id. Nothing is applied if any answer is refused. Pass idempotency_key to make a retry safe. A write tool: listed only when NetConverter enables the write tools for your tenant, and never in strict view. | Yes | Write | When enabled |
Account
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
get_accountMy account and usage | Your MCP access: tier, sign-in kind, the view mode in force and what each mode means, how many tools your session has, the per-minute and daily request limits, and today's tool calls and attributed cost with what remains. Takes no arguments. | No | Read | Default and strict |
Operations catalog
| Tool | What it answers | Needs a job | Kind | Listed |
|---|---|---|---|---|
list_operationsMigration operations catalog | The migration operations you can select (cleanup, policy and routing changes). With job_id set to a prescan, or to a conversion built from one, each row shows whether it applies and whether it was decided. Without a job, pass source_vendor, target_vendor and device_type. | Optional | Read | Default view |
get_decision_documentDecision document (YAML v2) | The prescan's decision document: the current revision or revision N, its sha256, the revision list and the operations ledger. These are your decisions, not your configuration. | Yes | Read | Default view |
validate_decision_documentValidate decision document | Validates a stored revision exactly as the worker will and returns line-numbered errors. | Yes | Read | Default view |
propose_operationsPropose operations (new revision) | Write tool, off unless enabled for your tenant. Applies 1 to 50 catalogued changes to the decision document and stores a new revision. Each change is {operation, parameters}: operation is the operation_id of a list_operations row (operation_id is also accepted as the key), and parameters matches that row's parameters schema. Pass base_sha256 so a concurrent edit is refused. | Yes | Write | When enabled |
submit_conversionSubmit conversion from prescan | Write tool, off unless enabled for your tenant. Queues a conversion from the prescan using a decision-document revision. Refuses with decision_required, queuing nothing, while a decision is still open. | Yes | Write | When enabled |
Not available
No tool returns raw configuration text, changes or deletes an existing job, or connects to your devices. The hosted server cannot read files from your computer; the local bridge submits a selected file instead.
Source and release context: MCP server · Trust and AI data-use statement