AgentRisk MCP server
Commit Sep 28, 2026Enforces risk guardrails for AI trading agents by analyzing portfolio exposure and validating proposed trades against custom YAML policies.
Also known as trycoin-ai/agentrisk
- Stars
- 3
- Tools
- —
- Downloads, 30 days
- 210
- Latest version
- 0.1.3
- Last commit
- Sep 28, 2026
- License
- MIT
What is AgentRisk MCP?
AgentRisk enforces risk guardrails for AI trading agents by analyzing portfolio exposure and validating proposed trades against custom YAML policies.
Summary generated from the project's README
About AgentRisk MCP server
AgentRisk provides a deterministic layer of protection for AI trading agents by validating proposed trades against user-defined YAML policies. It functions as a guardrail that sits in front of trade execution to ensure compliance with specific risk parameters. Developers can use this server locally via MCP clients like Claude Desktop or Claude Code, or integrate it directly into Python applications using the provided library. The system requires the uv tool for installation via uvx.
The server offers three primary tools to manage risk. The analyze_portfolio_risk tool generates reports on concentration and exposure, while check_trade_risk evaluates individual trades to return a PASS, WARN, or BLOCK verdict. Users can also manage their rules through the generate_risk_policy tool. It focuses on enforcing concentration caps, asset-class restrictions, and order sanity checks like maximum size or minimum cash floors. Because it is designed to fail-closed, any invalid input or missing policy will result in a blocked trade rather than a silent pass.
What you can do with it
- 1Enforce asset class restrictionsPrevent the agent from trading specific instruments like crypto or options.
Written from the project's README. Every use case is checked against a quote from it.
How to install AgentRisk MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall AgentRisk MCP in Claude Code
$ claude mcp add agentrisk -- uvx agentriskInstall AgentRisk MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"agentrisk": {
"command": "uvx",
"args": [
"agentrisk"
]
}
}
}Install AgentRisk MCP in VS Code
.vscode/mcp.json
{
"servers": {
"agentrisk": {
"type": "stdio",
"command": "uvx",
"args": [
"agentrisk"
]
}
}
}Install AgentRisk MCP in VS Code CLI
$ code --add-mcp '{"name":"agentrisk","type":"stdio","command":"uvx","args":["agentrisk"]}'Install AgentRisk MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"agentrisk": {
"command": "uvx",
"args": [
"agentrisk"
]
}
}
}Install AgentRisk MCP in Codex
$ codex mcp add agentrisk -- uvx agentriskInstall AgentRisk MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.agentrisk] command = "uvx" args = ["agentrisk"]
Install AgentRisk MCP in Gemini CLI
$ gemini mcp add -s user agentrisk uvx -- agentriskInstall AgentRisk MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"agentrisk": {
"command": "uvx",
"args": [
"agentrisk"
]
}
}
}Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachabletrycoin-ai/agentrisk is public and reachablegithub.com/trycoin-ai/agentrisk ↗Pass
- Package names this repository as its sourceagentrisk names trycoin-ai/agentrisk as sourcepypi.org/project/agentrisk/ ↗Pass
- Build provenance (Sigstore attestation)Sigstore attestation for agentriskpypi.org/project/agentrisk/ ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-09-28github.com/trycoin-ai/agentrisk ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageagentrisk looks like @4thtech-io/agentlinkFinding
- No critical advisory in the package or its dependencies7 resolved dependencies checked; no critical advisoryosv.dev ↗Pass
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubAgentRisk
Risk guardrails for AI trading agents. Your agent proposes; your policy decides.

AgentRisk sits in front of trade execution: it analyzes portfolio risk, checks each proposed trade against a policy you write, and manages that policy. The core is deterministic (same inputs, same verdict), transparent (policies are plain YAML you can read and edit), and fail-closed (invalid input or a missing policy blocks, never a silent pass).
AgentRisk never recommends trades and never executes them. A
PASSmeans a trade did not break the rules you wrote, not that it is safe or profitable. See DISCLAIMER.md.
Quickstart
The fastest path for a Claude agent is the Claude Code plugin. It installs the Agent
Skill and registers the MCP server in one step
(via uvx, so there is nothing else to install):
/plugin marketplace add trycoin-ai/agentrisk
/plugin install agentrisk@agentrisk
For any other MCP client, register the server yourself:
{
"mcpServers": {
"agentrisk": {
"command": "uvx",
"args": ["--from", "agentrisk[mcp]", "agentrisk-mcp"]
}
}
}
Then ask questions in plain English and the agent translates them into three tool calls:
| Tool | Question it answers | Returns |
|---|---|---|
analyze_portfolio_risk |
What risk am I holding? | Concentration, exposure, and policy-compliance report |
check_trade_risk |
Should this trade go through? | PASS / WARN / BLOCK with a one-line reason |
generate_risk_policy |
What are my rules? | A human-readable YAML policy (create, update, show) |
The plugin path needs uv for uvx. See
examples/claude_desktop.md for a two-minute Claude
Desktop setup.
Agent Skill
For Claude agents, AgentRisk ships an optional Agent Skill that encodes the
discipline the guardrail depends on: before any order reaches a broker, classify the
trade, call check_trade_risk, respect the verdict, and record the result, in that
order. The plugin above installs it with the server. To install just the skill:
- Claude Code or Claude Desktop: copy
skills/agentrisk/into~/.claude/skills/. - claude.ai: upload
skills/agentrisk/SKILL.mdas a skill.
The enforcement contract
AgentRisk returns advice. It cannot physically stop an order, so your integration must gate execution on the verdict:
result = check_trade_risk(portfolio, trade)
if not result.proceed:
refuse(result.summary) # BLOCK: never call the broker
elif result.acknowledgements_required:
confirm_with_user(result) # WARN: surface warnings first
else:
execute(trade) # PASS
If you call the broker regardless of the verdict, you have a logger, not a guardrail. See the integration guide.
Using the library directly
The three tools are also a pure Python library, with no network calls and no LLM in the core:
pip install agentrisk
Every parameter, return field, and error case is in the tool reference.
What it checks
- Concentration caps on single names, sectors, themes, and asset classes, evaluated on the simulated post-trade portfolio.
- Asset-class rules (allow, warn, or block) for crypto, options, and margin.
- Order sanity: max order size, minimum cash floor, insufficient-funds detection.
- Restricted symbols and data quality (stale snapshots warn, invalid ones block).
The policy reference covers every field and the safety behaviors: fail closed, exits are never trapped, only breach-worsening trades block, and the one-time bypass.
Documentation
| Doc | Contents |
|---|---|
| Concepts | The mental model and the agent/AgentRisk division of labor |
| Architecture | The module layout and how a call flows through the core |
| Policy reference | Every policy field and the check behavior it drives |
| Tool reference | Parameters, outputs, and error cases for all three tools |
| CLI reference | The agentrisk command: policy, check, analyze |
| Integration guide | The enforcement contract, broker MCP pairing, audit log |
| Threat model | What AgentRisk can and cannot protect against |
| Classification data | The open taxonomy and how to contribute corrections |
Roadmap
Next is deterministic stress scenarios, ETF look-through, and behavioral limits (v0.2), then options analytics, broker snapshot adapters, short positions, and multi-currency (v0.3). Trade recommendations, signals, execution, and telemetry are permanently out of scope. See the milestones.
Contributing
Classification-data corrections and new deterministic checks are especially welcome. See CONTRIBUTING.md, and keep the core pure: no network, no LLM calls, no hidden state.
License
MIT. See LICENSE.
AgentRisk MCP questions
How do I add AgentRisk MCP to Claude Code?
Run claude mcp add agentrisk -- uvx agentrisk in your terminal. Add --scope user to make it available in every project.
How do I add AgentRisk MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
Does AgentRisk execute trades?
No, AgentRisk never recommends or executes trades; it only provides a verdict that your integration must use to gate execution.
What happens if a policy is missing or input is invalid?
The system is designed to be fail-closed, meaning invalid inputs or missing policies will block the action rather than allowing it to pass silently.