What leaves your machine
Hooks never open a socket. The detached worker performs licensing and configuration requests outside the tool admission path. Free sessions without a license key make no licensing fetch.
The network inventory
Section titled “The network inventory”This is the inventory from the security page. The default destination is https://agentguard.run.
| Destination | What is sent | When and why |
|---|---|---|
POST /api/license/validate |
License key. | The worker checks license eligibility at session start or explicit activation. |
POST /api/license/seats |
license_key, machine_fingerprint, process_id, org_policy_sha256 or null. |
Registration, then a heartbeat about every five minutes while the session is live. |
GET /api/org/policy |
License key in the Authorization Bearer header. No request body. | Separate configuration lookup at session start and every fifth heartbeat. Eligible accounts receive the admin-published policy; Solo and unpublished policies receive no content. |
The two POSTs carry identifiers only; the policy GET is a third endpoint. The license key is a credential and should be kept private. HTTP connections also expose ordinary connection metadata such as source IP.
sequenceDiagram participant W as Detached worker participant A as agentguard.run Note over W: Hooks use local state only W->>A: POST /api/license/validate<br/>License key W->>A: POST /api/license/seats Note over W,A: license_key, machine_fingerprint,<br/>process_id, org_policy_sha256 or null W->>A: GET /api/org/policy<br/>Authorization Bearer license key Note over W,A: No request body for the policy GET
An operator can override only the license-validation base URL with AGENTGUARD_LICENSE_ENDPOINT; seat and org-policy endpoints stay fixed. A matching policy hash reports a loaded version; it does not attest enforcement.
What the plugin never sends
Section titled “What the plugin never sends”Tool arguments and output, prompts, file content, provider keys, signing keys, signed decisions, input hashes and byte counts are not sent by these requests. Guard Pack matching uses arguments in memory and records rule IDs, not the arguments.
The heartbeat contains no decision totals, block totals, fail-open totals, tool names, input, output, prompt, file content, signed receipt, input hash or output hash. Local usage and health records remain on the machine.
The dashboard stores the org policy your admin publishes, machine labels, revocation state and invite addresses. Local policy files are not uploaded. The policy uses a strict field allowlist; custom prompt or document fields are rejected.
The plugin policy reference describes this control-plane boundary:
AgentGuard’s server is a control plane for policy, not data. It stores the policy your admin writes, which seats are licensed, and which policy version each seat last reported. It never receives a tool call, a prompt, a file, or a receipt.
That statement describes the plugin’s traffic. The hosted tools below are separate submissions you choose to make. Installation can download packages from registries. Your host still calls its model provider, and an allowed tool can make its own requests. This inventory describes the plugin’s runtime, not every program on the machine.
If a request fails
Section titled “If a request fails”License, seat and org-policy failures select shadow mode with a reason. An AgentGuard service outage cannot deny the developer’s tool call; configured policy can block only while enforcement is active. The last good org policy copy is retained after a failed refresh, but retaining a copy does not mean enforcement is active.
Fail-open means loss of protection can allow an action. It is an availability choice, not evidence that the action was safe. Host permission checks still apply. Use the Getting started status check to inspect the effective mode and reason.
The separate hosted MCP
Section titled “The separate hosted MCP”agentguard.run/api/mcp uses MCP Streamable HTTP. Its three public tools are free and need no sign-in or license key. This hosted server never sees the plugin’s traffic.
| Tool | What you submit | What it does |
|---|---|---|
verify_receipt |
Receipt or decision-chain JSON, with a verification public key when needed. | Returns the public verifier’s cryptographic verdict. A valid signature does not establish that a recorded statement is true. |
start_trial |
An email address you authorize. | Uses the website’s setup-email flow. Free Enforce needs no key; this sends the setup steps, as described in skill.md. |
ask_docs |
A question about AgentGuard. | Uses the website’s fixed support answers first, then its guarded help model for other questions. |
Every tool is rate limited per IP. There are no other tools or MCP resources. The server does not provide account, billing, license lookup or customer-data access.
The receipt, question or email you submit to a hosted tool is sent to AgentGuard’s server. A question outside the fixed answers may be sent to the guarded help model or forwarded to support through the existing escalation flow. Do not include customer information, prompts, private files, credentials or signing keys in tool arguments. For host-specific connection instructions, see Hosted MCP setup.