Agent Q MCP server
io.github.stelis-dev/agent-q·v0.2.2Local MCP server for requesting signatures from Agent-Q Firmware.
0 stars38 downloads/wk
Reviews
Write oneNobody has reviewed Agent Q yet.
If you have run it, two minutes of your experience saves the next person an afternoon.
Agent Q tools (15, 1 write)
write = sends, deletes, buys or postsRead from the package source without running it. The installed server may list more.
connect_deviceOpen a communication session with a known Agent-Q Firmware device. Resolves the target device by deviceId, by purpose, or by the default active device. Sends a connect request that requires Firmware-owned device-local approval. Writes a status handshake to candidate USB serial ports while locating the device. Connect is not signing approval and does not authorize signing. Session is held in Agent-
disconnect_deviceEnd a previously approved Agent-Q Firmware session. Resolves the target device by deviceId, by purpose, or by the default active device. Returns 'not_connected' without contacting Firmware when there is no Agent-Q runtime session. Writes a status handshake to candidate USB serial ports when locating the device. Disconnect does not require physical approval.
get_accountsList the public accounts (chain, address, public key) held by a provisioned Agent-Q Firmware device over an approved session. Resolves the target device by deviceId, by purpose, or by the default active device. Requires a prior connect_device approval; returns 'not_connected' without contacting Firmware when there is no Agent-Q runtime session. Read-only: no signing, no private material, and no se
get_approval_historyRead a bounded page of Firmware-owned approval history over an approved session. This is not on-chain history and is read-only: no raw requests, session ids, private material, PINs, client names, or full policy documents are returned.
get_capabilitiesRead Firmware-authored supported chains, public account schemes, and supported signing methods over an approved session. Resolves the target device by deviceId, by purpose, or by the default active device. Requires a prior connect_device approval; returns 'not_connected' without contacting Firmware when there is no Agent-Q runtime session. signing.authorization is Firmware-authored read-only state
get_device_statusRead live or cached status for a known Agent-Q Firmware device by writing a status handshake to candidate USB serial ports. Resolves the device by deviceId, by purpose, or by the default active device.
identify_devicesAsk discovered Agent-Q Firmware devices to display short identification codes. Writes a status handshake to candidate USB serial ports before sending the identify request.
list_devicesList Agent-Q Firmware devices known to Agent-Q, including local label, purpose routing assignments, and any in-memory connection session metadata. Reads from local Agent-Q process state only; does not contact Firmware.
policy_getRead the Firmware-owned active policy document over an approved session. Resolves the target device by deviceId, by purpose, or by the default active device. Requires a prior connect_device approval; returns 'not_connected' without contacting Firmware when there is no Agent-Q runtime session. Read-only: no policy update, no signing, no private material, and no session id is ever returned.
policy_proposewrite actionSubmit a bounded current-schema active-policy proposal to Agent-Q Firmware. Firmware validates the proposal, shows a device-local policy summary review, starts local PIN approval only after device-local Continue, and returns the terminal policy proposal outcome. This is a request path only: Agent-Q and MCP do not store, apply, or decide policy.
scan_devicesFind USB-connected Agent-Q Firmware devices by writing a status handshake to candidate USB serial ports, and report sanitized candidate failure reasons.
select_deviceSet a previously discovered Agent-Q Firmware device as the default active device, or as the active device for a named routing purpose. 'purpose' is local Agent-Q process routing metadata, not security policy. Updates local Agent-Q process state only; does not contact Firmware.
set_device_metadataSet local metadata for a known Agent-Q Firmware device. 'label' is human-readable local metadata, not a security boundary and not device authority. Pass null to clear the label. Updates local Agent-Q process state only; does not contact Firmware.
sign_personal_messageRequest a Firmware-owned Sui personal-message signature over an approved session. Agent-Q and MCP do not store keys, choose authorization, or make signing decisions. The current Firmware implementation signs this method only in user-confirmed authorization mode; policy authorization mode fails closed.
sign_transactionRequest a Firmware-owned transaction signature over an approved session. Agent-Q and MCP do not store keys, choose authorization, or make signing decisions; Firmware uses its local signing authorization mode to select policy authorization or user confirmation and returns signing outcome.
Public scan report
scanner v0.1.9 · 2026-09-26 · same rubric, same numbers if you re-run it
- Code scan17 source files scanned25/25
- –Live reliabilityno gateway calls yet and no remote to proben/a
- –Tool poisoningtools not inspected (local package is not executed); not countedn/a
- Auth qualitylocal package, no credentials required12/15
- Maintenancelast push 54 days ago12/15
- Maintainer identityregistry namespace matches repository owner7/10
How Agent Q compares
Against every graded AI and agents MCP server, from our own scans and uptime checks. Updated after each scan run.
- Safety score
- 86/100
- higher than 75% of 4,460 AI and agents servers
- Tools
- 15
- the typical AI and agents server has 6
- Tools that change things
- 7%
- typical: 0%. Fewer write tools means less an agent can break
- Downloads a week
- 38
- more than 30% of 1,538 AI and agents servers with a package
What its tools let an agent do: 13 read · 2 change
Install Agent Q in Claude Code, Cursor or VS Code
Runs npx -y @stelis/agent-q on your machine. Read the scan report first; the gateway never runs local packages.
claude mcp add agent-q -- npx -y @stelis/agent-q
What the publisher says
From the Agent Q repository's README, as published. We do not edit it. Read it on GitHub
Agent-Q
Development status: Agent-Q is an active development project with
hardware-tested Sui signing paths for CLI, MCP, and supported provider flows.
The current StackChan CoreS3 Firmware path uses DEV_PROFILE material intended
for development and demos, not real-asset custody. See Current Status for
storage and profile limitations.
Concept
Agent-Q separates execution surfaces from signing authority.
Execution capability is not signing authority. Agent-Q keeps that boundary explicit for agent-driven and local Sui signing flows.
This README describes the Agent-Q authority-boundary architecture for local Sui signing flows. Current implementation status and profile limitations are tracked in docs/IMPLEMENTATION_STATUS.md.
MCP clients, Sui CLI, and supported Sui app adapters submit bounded requests through the local agent-q host process. AI agent runtimes usually reach Agent-Q through MCP clients. The host relays requests; it does not store signing material, decide policy, or approve signing.
Agent-Q Firmware runs on a separate signing device. It holds signing material and policy for the implemented material profile, checks device state and request contents, applies the implemented policy or device-local approval gate, and returns signed results or explicit failures.
Signing request transport stays local. Signing authority stays on the device. Requesters can ask for signatures, but they cannot hold signing material, choose the authorization mode, or approve their own requests.
Structure
Agent-Q has three roles:
submission, or status requests. MCP clients, Sui CLI, Sui app adapters, and the Admin Page are request surfaces, not signing authorities.
- requesters create signing, account, policy read, policy proposal
tools, the local HTTP API, the Admin Page, and adapter-facing request surfaces, then relays bounded requests to Firmware.
- the host process is the local agent-q process. It exposes MCP stdio
owns policy checks, device-local approval gates, signing, persistence, and cleanup, and holds signing material for the implemented material profile.
- Agent-Q Firmware is the software running on the signing device. Firmware
Agent-Q uses a simple authority model:
state.
- requesters create requests;
- the host process relays requests;
- Firmware handles each implemented request type according to device-local
For signing requests, Firmware decides whether the implemented policy or device-local approval gate permits signing.
Unsupported chains and methods fail explicitly.
Use Agent-Q
See packages/agent-q/README.md.
- Sui users can register agent-q-sui-signer as a Sui CLI external signer.
packages/agent-q/README.md.
- MCP clients can run the local server and call Agent-Q MCP tools. See
Standard adapter flows. See packages/provider-sui/README.md and packages/sample-sui-dapp-kit/README.md.
- Sui app developers can use the Sui provider package for supported Wallet
non-production test web sample. See packages/sample-zklogin-test-web/README.md.
- Sui zkLogin browser setup and sign-only transaction checks live in the
local approval, and bounded signing flows should start with firmware/README.md.
- Firmware and hardware developers working on device-owned signing policy,
Packages
Shortened. The full README is on GitHub.
Nothing above is checked by us. What we check is on the safety report.
Agent Q: common questions
- Is Agent Q MCP server safe?
- Yes, by our scan: it is graded A (86/100). Read the Agent Q safety report
- How do I install Agent Q?
- It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.
- Does Agent Q need an API key?
- Not as far as the registry entry and our scan can tell: no credentials are declared or required.
- Is Agent Q maintained?
- The last commit was 92 days ago (2026-08-04). The latest release is v0.2.2.
- How does Agent Q compare with other AI and agents MCP servers?
- safety score 86/100 (higher than 75% of 4,460 AI and agents servers); tools 15 (the typical AI and agents server has 6); tools that change things 7% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.