API Documentation

SDK Python SDK

The easiest way to integrate AgentShield. Install the official Python package:

pip install agentshield-ai

Basic Usage

from agentshield import AgentShield

shield = AgentShield(api_key="ask_your_key_here")

# Send an event for analysis
result = shield.track(
    agent_name="my-chatbot",
    user_input="Do you offer free shipping?",
    agent_output="Yes, we offer free shipping on orders over $50.",
    action_taken="respond",
)

if result["alert_triggered"]:
    print(f"ALERT: {result['alert_reason']}")

Decorator (Auto-Monitor)

Wrap any agent function to automatically track every call:

from agentshield import AgentShield

shield = AgentShield(api_key="ask_your_key_here")

@shield.monitor("customer-support")
def my_agent(question):
    # Your LLM call here
    return call_llm(question)

# Every call is now monitored automatically
response = my_agent("Can I get a refund?")

Get Stats

stats = shield.stats()
print(f"Total events: {stats['total_events']}")
print(f"Alert rate: {stats['alert_rate']}%")

Tracing

with shield.start_trace("support-bot") as trace:
    trace.span("llm_call", "call_gpt4",
               input_text=prompt, output_text=response,
               tokens_input=150, tokens_output=300,
               model_used="gpt-4")
    trace.span("tool_call", "search_database",
               input_text="query", output_text="results")

Approvals

approval_id = shield.request_approval(
    agent_name="billing-bot",
    action_description="Process refund of $2,500",
    estimated_value=2500.0
)

result = shield.wait_for_approval(approval_id, timeout=300)
if result["status"] == "approved":
    process_refund()

SDK Reference

Method	Description
shield.track(...)	Send a single event for risk analysis. Accepts optional: trace_id, session_id, tokens_input, tokens_output, model_used, estimated_cost.
@shield.monitor(name)	Decorator that auto-tracks every call to your agent function.
shield.stats()	Get aggregated statistics (total events, alerts, risk breakdown).
shield.start_trace(agent_name)	Returns a Trace context manager. Use `with shield.start_trace("bot") as trace:` then call `trace.span(...)`.
trace.span(span_type, name, ...)	Add a span to a trace. Accepts: input_text, output_text, tokens_input, tokens_output, model_used.
shield.request_approval(...)	Request human approval. Args: agent_name, action_description, estimated_value. Returns approval_id string.
shield.wait_for_approval(id, ...)	Poll for approval decision. Args: approval_id, timeout=300, poll_interval=5. Returns status dict.

View on PyPI: pypi.org/project/agentshield-ai

SDK Framework Integrations

AgentShield integrates with popular AI agent frameworks. Every agent execution, tool call, and LLM call is automatically traced — zero code changes to your agents.

LANGCHAIN

LangChain / LangGraph

Automatic tracing for LangChain chains and agents via a callback handler. Captures every LLM call, tool use, chain step, and retrieval.

pip install agentshield-ai[langchain]

from agentshield import AgentShield
from agentshield.langchain_callback import AgentShieldCallbackHandler

shield = AgentShield(api_key="ask_your_key_here")
callback = AgentShieldCallbackHandler(shield, agent_name="my-chain")

# Pass to any LangChain chain or agent
chain = prompt | llm | parser
result = chain.invoke({"input": "Hello"}, config={"callbacks": [callback]})

What's captured: LLM calls (model, tokens, cost), tool calls (name, args, result), chain steps, retrieval queries, errors — all with parent-child relationships.

CREWAI

CrewAI

Automatic tracing for CrewAI crews via an event listener. Captures crew lifecycle, agent executions, task results, tool calls, and LLM calls.

pip install agentshield-ai[crewai]

from agentshield import AgentShield
from agentshield.crewai_listener import AgentShieldCrewAIListener

shield = AgentShield(api_key="ask_your_key_here")
listener = AgentShieldCrewAIListener(shield, agent_name="my-crew")

# That's it! Now run any CrewAI crew — everything is traced automatically.
crew = Crew(agents=[researcher, writer], tasks=[research_task, write_task])
crew.kickoff()

What's captured: Crew kickoff/completion, agent executions (role, goal), task lifecycle, tool calls (name, args, cache hits), LLM calls (model, tokens, response) — all with parent-child span tree.

OPENAI

OpenAI Agents SDK

Automatic tracing for OpenAI Agents via a TracingProcessor. Captures agent executions, LLM generations, tool calls, handoffs, and guardrails.

pip install agentshield-ai openai-agents

from agentshield import AgentShield
from agentshield.openai_agents_tracer import AgentShieldOpenAIAgentsTracer

shield = AgentShield(api_key="ask_your_key_here")
tracer = AgentShieldOpenAIAgentsTracer(shield, agent_name="my-agent")

# That's it! All agent activity is traced automatically.
from agents import Agent, Runner
agent = Agent(name="Bot", instructions="Be helpful.")
result = Runner.run_sync(agent, "Hello!")

What's captured: Agent executions, LLM generations (model, tokens, cost), tool/function calls, agent handoffs, guardrail evaluations — all with parent-child span tree.

MCP

Model Context Protocol (MCP)

NEW

Zero-code integration for MCP-compatible AI agents. Works with Claude Desktop, Cursor, Windsurf, Cline, and any tool that supports the Model Context Protocol. No SDK code needed — just add a JSON config.

1. Install

pip install agentshield-ai[mcp]

2. Add to your MCP config

Claude Desktop: ~/.claude/claude_desktop_config.json | Cursor: .cursor/mcp.json

{
  "mcpServers": {
    "agentshield": {
      "command": "python",
      "args": ["-m", "agentshield.mcp_server"],
      "env": {
        "AGENTSHIELD_API_KEY": "ask_your_key_here"
      }
    }
  }
}

3. That's it!

Your AI agent now has access to AgentShield tools: track_event, start_trace, add_span, end_trace, request_approval, check_approval, get_stats, get_cost_summary.

Available tools: track_event (risk analysis), start_trace / add_span / end_trace (execution tracing), request_approval / check_approval (human-in-the-loop), get_stats, get_cost_summary.

Install all integrations at once: pip install agentshield-ai[all]

Authentication

All API requests require an API key sent in the X-API-Key header.

Your API key is available on your Dashboard after signing up. Keys follow the format ask_...

curl -H "X-API-Key: ask_your_key_here" \
  https://useagentshield.net/api/events/stats

MCPS Identity Verification

AgentShield supports MCPS (MCP Secure) — a cryptographic identity verification protocol for AI agents using ECDSA P-256 signatures. Agents with verified MCPS passports get improved risk scoring, trust indicators in the dashboard, and adaptive alert thresholds — higher trust agents receive fewer false positives while anomalies are escalated faster.

MCPS uses trust levels from L0 (unsigned) to L4 (fully audited). Higher trust levels reduce the risk score and adjust alert sensitivity:

Level	Name	Description
L0	Unsigned	No passport attached
L1	Identified	Passport present, signature not verified
L2	Verified	ECDSA signature verified
L3	Scanned	Verified + security scan passed
L4	Audited	Verified + full audit trail

Alert Thresholds by Trust Level

Trust Level	Minimum Alert Level	Behavior
L0-L1	Low (all alerts fire)	Full monitoring — unknown agent
L2	Medium (low suppressed)	Reduced noise — verified identity
L3-L4	High (low + medium suppressed)	Anomaly detection — trusted agent

Below-threshold alerts are suppressed automatically. The risk level is still recorded for scoring and reporting, but no alert is triggered — reducing dashboard noise while maintaining full observability.

Usage with SDK

from agentshield import AgentShield

shield = AgentShield(
    api_key="ask_your_key",
    mcps_passport={
        "agent_id": "your-agent-uuid",
        "trust_level": "L2",
        "signature": "base64-ecdsa-signature",
        "timestamp": "2026-03-14T10:00:00Z",
        "issuer": "mcps-authority-id",
    }
)
# All API calls now include the MCPS passport header

Raw HTTP Header

X-MCPS-Passport: {"agent_id": "...", "trust_level": "L2", "signature": "...", "timestamp": "..."}

MCPS protocol by CyberSecAI Ltd (MIT License).

Base URL

https://useagentshield.net/api

POST /api/events

Send an event every time your AI agent processes a request. AgentShield will analyze it in real-time and return a risk assessment.

Request Body

Field	Type	Required	Description
agent_name	string	Yes	Unique name for your agent (e.g. "customer-support-bot")
event_type	string	No	"response" (default), "action", or "error"
user_input	string	No	What the user asked your agent
agent_output	string	No	What your agent responded — this is analyzed for risks
action_taken	string	No	Action the agent performed (e.g. "apply_discount", "process_refund")
value	float	No	Monetary value involved (used for anomaly detection)
metadata	object	No	Any additional data you want to attach
trace_id	string	No	Link this event to a trace
session_id	string	No	Group events into a session
tokens_input	int	No	Input tokens used
tokens_output	int	No	Output tokens used
model_used	string	No	LLM model name (e.g. "gpt-4", "claude-3")
estimated_cost	float	No	Estimated cost in USD

Response

Field	Type	Description
event_id	int	Unique event ID
risk_level	string	"low", "medium", "high", or "critical"
alert_triggered	bool	Whether this event triggered an alert
alert_reason	string	Why the alert was triggered (null if no alert)

Example

REQUEST

curl -X POST https://useagentshield.net/api/events \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{
    "agent_name": "support-bot",
    "event_type": "response",
    "user_input": "Can I get a refund?",
    "agent_output": "I have processed your full refund of $500",
    "action_taken": "process_refund",
    "value": 500.00
  }'

RESPONSE

{
  "event_id": 42,
  "risk_level": "low",
  "alert_triggered": false,
  "alert_reason": null
}

GET /api/events/stats

Get aggregated statistics for all events sent by your agents.

Response

{
  "total_events": 1250,
  "total_alerts": 23,
  "alert_rate": 1.8,
  "risk_breakdown": {
    "low": 1100,
    "medium": 127,
    "high": 20,
    "critical": 3
  },
  "agents": [
    {"agent_name": "support-bot", "count": 800, "alerts": 15},
    {"agent_name": "sales-agent", "count": 450, "alerts": 8}
  ]
}

Agent Tracing Starter+

Capture end-to-end traces of your agent's execution, including LLM calls, tool usage, and sub-agent handoffs.

POST /api/traces

Start a new trace for an agent execution.

Request Body

Field	Type	Required	Description
agent_name	string	Yes	Name of the agent being traced
session_id	string	No	Group traces into a session
metadata	object	No	Additional context for this trace

Response

Field	Type	Description
trace_id	string	Unique trace identifier
status	string	"active"
created_at	string	ISO 8601 timestamp

Example

curl -X POST https://useagentshield.net/api/traces \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{"agent_name": "support-bot", "session_id": "sess_abc123"}'

PUT /api/traces/{trace_id}

Complete or mark a trace as errored.

Request Body

Field	Type	Required	Description
status	string	Yes	"completed" or "error"

curl -X PUT https://useagentshield.net/api/traces/trc_abc123 \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{"status": "completed"}'

GET /api/traces

List traces with optional filters.

Query Parameters

Param	Type	Description
page	int	Page number (default: 1)
per_page	int	Results per page (default: 20)
agent_name	string	Filter by agent name
risk_level	string	Filter by risk level
session_id	string	Filter by session

Response

{
  "traces": [...],
  "total": 150,
  "page": 1,
  "per_page": 20
}

GET /api/traces/{trace_id}

Get a single trace with its nested spans tree.

Response

{
  "trace": { "trace_id": "trc_abc123", "agent_name": "support-bot", "status": "completed", ... },
  "spans": [
    { "span_id": "spn_1", "span_type": "llm_call", "name": "call_gpt4", "children": [...] }
  ]
}

POST /api/spans

Add a span to an active trace.

Request Body

Field	Type	Required	Description
trace_id	string	Yes	Parent trace ID
span_type	string	Yes	llm_call, tool_call, retrieval, function, agent_handoff, or custom
name	string	Yes	Descriptive name for this span
parent_span_id	string	No	Nest under a parent span
input_text	string	No	Input to this span
output_text	string	No	Output from this span
tokens_input	int	No	Input tokens used
tokens_output	int	No	Output tokens used
model_used	string	No	LLM model name
estimated_cost	float	No	Estimated cost in USD
duration_ms	int	No	Duration in milliseconds
metadata	object	No	Additional span data

curl -X POST https://useagentshield.net/api/spans \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{
    "trace_id": "trc_abc123",
    "span_type": "llm_call",
    "name": "call_gpt4",
    "input_text": "What is our refund policy?",
    "output_text": "Our refund policy allows...",
    "tokens_input": 150,
    "tokens_output": 300,
    "model_used": "gpt-4"
  }'

PUT /api/spans/{span_id}

Update or complete an existing span.

Request Body

Field	Type	Description
output_text	string	Output from this span
status	string	"completed" or "error"
duration_ms	int	Duration in milliseconds
tokens_input	int	Input tokens used
tokens_output	int	Output tokens used
estimated_cost	float	Estimated cost in USD

curl -X PUT https://useagentshield.net/api/spans/spn_xyz789 \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{"status": "completed", "duration_ms": 1250}'

Human-in-the-Loop Approvals Starter+

Pause agent execution and require human approval before high-risk actions proceed.

POST /api/approvals

Request human approval for an agent action.

Request Body

Field	Type	Required	Description
agent_name	string	Yes	Name of the requesting agent
action_description	string	Yes	Human-readable description of the action
trace_id	string	No	Link to a trace
risk_level	string	No	Risk level override
estimated_value	float	No	Monetary value of the action

Response

Field	Type	Description
approval_id	string	Unique approval request ID
status	string	"pending"
expires_at	string	ISO 8601 expiration timestamp

curl -X POST https://useagentshield.net/api/approvals \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{
    "agent_name": "billing-bot",
    "action_description": "Process refund of $2,500",
    "estimated_value": 2500.0
  }'

GET /api/approvals/{approval_id}

Check the status of an approval request.

Response

Field	Type	Description
approval_id	string	Unique approval ID
status	string	"pending", "approved", "rejected", or "expired"
decided_by	string	Email of the reviewer (null if pending)
decision_reason	string	Reason provided by the reviewer

curl https://useagentshield.net/api/approvals/apr_abc123 \
  -H "X-API-Key: ask_your_key_here"

POST /api/approvals/{approval_id}/decide

Approve or reject a pending request (typically called from the dashboard).

Request Body

Field	Type	Required	Description
decision	string	Yes	"approved" or "rejected"
reason	string	No	Reason for the decision

curl -X POST https://useagentshield.net/api/approvals/apr_abc123/decide \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{"decision": "approved", "reason": "Verified with customer"}'

GET /api/approvals

List all approval requests. Filter by status with ?status=pending.

curl https://useagentshield.net/api/approvals?status=pending \
  -H "X-API-Key: ask_your_key_here"

POST /api/approval-rules

Create an automatic approval rule.

Request Body

Field	Type	Description
agent_name	string	Agent this rule applies to
rule_type	string	risk_level, value_threshold, action_type, or always
rule_value	string	Value for the rule (e.g. "high", "1000")
auto_approve_after_minutes	int	Auto-approve if no decision after N minutes

curl -X POST https://useagentshield.net/api/approval-rules \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{"agent_name": "billing-bot", "rule_type": "value_threshold", "rule_value": "1000"}'

GET /api/approval-rules

List all approval rules.

curl https://useagentshield.net/api/approval-rules \
  -H "X-API-Key: ask_your_key_here"

DELETE /api/approval-rules/{id}

Delete an approval rule.

curl -X DELETE https://useagentshield.net/api/approval-rules/rule_abc123 \
  -H "X-API-Key: ask_your_key_here"

Cost Attribution Starter+

Track LLM costs per agent, model, and time period. Set budget alerts to avoid surprise bills.

GET /api/costs/summary

Get cost summary for the current billing month.

Response

Field	Type	Description
total_cost	float	Total cost in USD this month
total_tokens_input	int	Total input tokens
total_tokens_output	int	Total output tokens
total_traces	int	Total traces this month
avg_cost_per_trace	float	Average cost per trace

curl https://useagentshield.net/api/costs/summary \
  -H "X-API-Key: ask_your_key_here"

GET /api/costs/by-agent

Get cost breakdown grouped by agent.

Response

[
  { "agent_name": "support-bot", "total_cost": 45.20, "total_tokens": 1500000, "trace_count": 320 },
  { "agent_name": "billing-bot", "total_cost": 12.80, "total_tokens": 420000, "trace_count": 85 }
]

GET /api/costs/by-model

Get cost breakdown grouped by LLM model.

Response

[
  { "model": "gpt-4", "total_cost": 38.50, "total_tokens": 800000, "span_count": 250 },
  { "model": "claude-3", "total_cost": 19.50, "total_tokens": 1100000, "span_count": 155 }
]

GET /api/costs/daily

Get daily cost trend for the last 30 days.

Response

[
  { "date": "2026-03-05", "total_cost": 2.40, "trace_count": 15 },
  { "date": "2026-03-04", "total_cost": 3.10, "trace_count": 22 },
  ...
]

POST /api/costs/budgets

Create a budget alert for an agent.

Request Body

Field	Type	Description
agent_name	string	Agent to set budget for
budget_type	string	"daily", "weekly", or "monthly"
budget_amount	float	Budget limit in USD
alert_at_percent	int	Alert when this % of budget is used (e.g. 80)

curl -X POST https://useagentshield.net/api/costs/budgets \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{"agent_name": "support-bot", "budget_type": "monthly", "budget_amount": 100.0, "alert_at_percent": 80}'

GET /api/costs/budgets

List all budget alerts.

curl https://useagentshield.net/api/costs/budgets \
  -H "X-API-Key: ask_your_key_here"

DELETE /api/costs/budgets/{id}

Delete a budget alert.

curl -X DELETE https://useagentshield.net/api/costs/budgets/bgt_abc123 \
  -H "X-API-Key: ask_your_key_here"

Agent Risk Score Starter+ NEW

Continuous risk score (0-1000) per agent based on alert rate, error rate, cost stability, approval compliance, and more. Higher score = safer agent.

GET /api/risk-score?agent_name=X

Get risk score for a specific agent. Recalculates on each request.

Response

{
  "agent_name": "support-bot",
  "risk_score": 720,
  "previous_score": 735,
  "score_change": -15,
  "band": "good",
  "color": "blue",
  "components": {
    "alert_rate": {"score": 900, "weight": 25},
    "risk_distribution": {"score": 800, "weight": 20},
    "hallucination_rate": {"score": 500, "weight": 15},
    ...
  }
}

GET /api/risk-score/all

Get risk scores for all agents.

[
  {"agent_name": "support-bot", "risk_score": 720, "band": "good", "score_change": -15},
  {"agent_name": "billing-bot", "risk_score": 450, "band": "fair", "score_change": 30}
]

SDK Usage

score = shield.get_risk_score("support-bot")
print(f"Score: {score['risk_score']}/1000 ({score['band']})")

Cost Prediction Starter+ NEW

Predict the cost of an agent task BEFORE running, based on historical traces. Returns p25/p50/p95 cost estimates with confidence.

GET /api/costs/predict?agent_name=X&model_used=Y

Get cost prediction for next agent execution. Requires 5+ completed traces.

Response

{
  "prediction_id": "prd_abc123...",
  "agent_name": "support-bot",
  "predicted_cost": {"low": 0.42, "mid": 1.23, "high": 4.87},
  "predicted_tokens": {"input": 1500, "output": 800},
  "confidence": 0.87,
  "based_on_traces": 26,
  "budget_recommendation": 5.84,
  "model_used": "gpt-4"
}

POST /api/costs/predict/track

Track actual cost against a prediction to measure accuracy.

// Request
{"prediction_id": "prd_abc123", "actual_cost": 1.45}

// Response
{"prediction_id": "prd_abc123", "predicted_mid": 1.23, "actual_cost": 1.45, "accuracy_pct": 0.82}

SDK Usage

prediction = shield.predict_cost("support-bot", model_used="gpt-4")
print(f"Estimated: ${prediction['predicted_cost']['mid']}")

Blast Radius Starter+ NEW

Estimate the maximum potential damage an agent can cause based on its permissions, history, and financial exposure. Get actionable mitigations.

GET /api/blast-radius?agent_name=X

Get full blast radius analysis for an agent.

Response

{
  "agent_name": "billing-bot",
  "blast_radius_score": 67,
  "blast_radius_band": "high",
  "estimated_max_damage_usd": 23000.00,
  "reversibility_score": 35,
  "scope": {
    "tool_types": ["llm_call", "tool_call"],
    "has_destructive_actions": true,
    "max_single_transaction": 2500.00
  },
  "risk_factors": [
    "Agent has executed destructive actions 12 times",
    "No approval rules for transactions up to $2,500.00"
  ],
  "mitigations": [
    "Add approval rules for financial actions",
    "Set approval rule for transactions > $1,250"
  ]
}

GET /api/blast-radius/all

Get blast radius summary for all agents.

[
  {"agent_name": "billing-bot", "blast_radius_score": 67, "blast_radius_band": "high", "estimated_max_damage_usd": 23000},
  {"agent_name": "support-bot", "blast_radius_score": 12, "blast_radius_band": "low", "estimated_max_damage_usd": 0}
]

SDK Usage

blast = shield.get_blast_radius("billing-bot")
print(f"Score: {blast['blast_radius_score']}/100 ({blast['blast_radius_band']})")
print(f"Max damage: ${blast['estimated_max_damage_usd']:,.0f}")

Pre-Production Testing Starter+

Run adversarial, bias, edge-case, and compliance tests against your agents before deploying to production.

POST /api/test-suites

Create a new test suite for an agent.

Request Body

Field	Type	Required	Description
name	string	Yes	Name for this test suite
agent_name	string	Yes	Agent to test
test_type	string	No	adversarial, bias, edge_case, compliance, or custom
custom_cases	array	No	Array of custom test case objects

Response

Field	Type	Description
suite_id	string	Unique test suite ID
test_cases	array	Generated test cases
status	string	"created"

curl -X POST https://useagentshield.net/api/test-suites \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{"name": "Refund safety tests", "agent_name": "billing-bot", "test_type": "adversarial"}'

GET /api/test-suites

List all test suites.

curl https://useagentshield.net/api/test-suites \
  -H "X-API-Key: ask_your_key_here"

GET /api/test-suites/{suite_id}

Get a test suite with its results.

curl https://useagentshield.net/api/test-suites/ts_abc123 \
  -H "X-API-Key: ask_your_key_here"

POST /api/test-suites/{suite_id}/results

Submit test results for a suite run.

curl -X POST https://useagentshield.net/api/test-suites/ts_abc123/results \
  -H "Content-Type: application/json" \
  -H "X-API-Key: ask_your_key_here" \
  -d '{"results": [{"case_id": "tc_1", "passed": true, "output": "..."}]}'

Compliance Reports Pro+

Generate compliance reports for EU AI Act, NIST AI RMF, and export full audit trails.

GET /api/reports/compliance

Generate a compliance report for a specific framework.

Query Parameters

Param	Type	Description
framework	string	eu_ai_act or nist_ai_rmf
period_days	int	Report period in days (default: 30)

curl "https://useagentshield.net/api/reports/compliance?framework=eu_ai_act&period_days=30" \
  -H "X-API-Key: ask_your_key_here"

GET /api/reports/audit-trail

Export a full audit trail of all agent activity.

Query Parameters

Param	Type	Description
format	string	json or csv

curl "https://useagentshield.net/api/reports/audit-trail?format=csv" \
  -H "X-API-Key: ask_your_key_here"

Error Codes

Code	Meaning
401	Invalid API key
403	Plan does not include this feature
429	Plan limit reached (agent limit or monthly event limit)
422	Invalid request body (missing required fields)

What We Detect

AgentShield analyzes your agent's output in real-time for these risk categories:

CRITICAL

• Discrimination (age, gender, race, disability)
• Zero/negative values in payments

HIGH

• Unauthorized promises (free forever, unlimited)
• Medical or legal advice
• Unauthorized refund/discount promises
• Unusually high transaction values

MEDIUM

• Agent errors
• High-value transactions (>$1,000)

LOW

• Normal agent behavior
• No risks detected

Plan Limits

Feature	Free ($0)	Starter ($49/mo)	Pro ($149/mo)	Enterprise
API + Dashboard	Yes	Yes	Yes	Yes
Agents	1	5	20	Unlimited
Events / month	1,000	50,000	500,000	Unlimited
Risk Analysis	Keyword	AI-Powered	AI-Powered	AI-Powered
Agent Tracing	—	10K/mo	100K/mo	Unlimited
Approvals	—	100/mo	1K/mo	Unlimited
Cost Tracking	—	Yes	Yes	Yes
Testing	—	10 runs/mo	100 runs/mo	Unlimited
Compliance Reports	—	—	Yes	Yes
Email Alerts	Yes	Yes	Yes	Yes
Webhooks / Slack	Yes	Yes	Yes	Yes

Code Examples

Python (SDK — Recommended)

from agentshield import AgentShield

shield = AgentShield(api_key="ask_your_key_here")

result = shield.track(
    agent_name="my-chatbot",
    user_input=user_message,
    agent_output=agent_response,
    action_taken="respond",
)

if result["alert_triggered"]:
    print(f"ALERT: {result['alert_reason']}")

Python (requests)

import requests

response = requests.post(
    "https://useagentshield.net/api/events",
    headers={"X-API-Key": "ask_your_key_here"},
    json={
        "agent_name": "my-chatbot",
        "user_input": user_message,
        "agent_output": agent_response,
        "action_taken": "respond",
    }
)
result = response.json()

JavaScript / Node.js

const response = await fetch("https://useagentshield.net/api/events", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "X-API-Key": "ask_your_key_here",
  },
  body: JSON.stringify({
    agent_name: "my-chatbot",
    user_input: userMessage,
    agent_output: agentResponse,
    action_taken: "respond",
  }),
});

const result = await response.json();
if (result.alert_triggered) {
  console.warn(`ALERT: ${result.alert_reason}`);
}

Quick Start

Contents

SDK Python SDK

Basic Usage

Decorator (Auto-Monitor)

Get Stats

Tracing

Approvals

SDK Reference

SDK Framework Integrations

LangChain / LangGraph

CrewAI

OpenAI Agents SDK

Model Context Protocol (MCP)

1. Install

2. Add to your MCP config

3. That's it!

Authentication

MCPS Identity Verification

Alert Thresholds by Trust Level

Usage with SDK

Raw HTTP Header

Base URL

POST /api/events

Request Body

Response

Example

GET /api/events/stats

Response

Agent Tracing Starter+

POST /api/traces

Request Body

Response

Example

PUT /api/traces/{trace_id}

Request Body

GET /api/traces

Query Parameters

Response

GET /api/traces/{trace_id}

Response

POST /api/spans

Request Body

PUT /api/spans/{span_id}

Request Body

Human-in-the-Loop Approvals Starter+

POST /api/approvals

Request Body

Response

GET /api/approvals/{approval_id}

Response

POST /api/approvals/{approval_id}/decide

Request Body

GET /api/approvals

POST /api/approval-rules

Request Body

GET /api/approval-rules

DELETE /api/approval-rules/{id}

Cost Attribution Starter+

GET /api/costs/summary

Response

GET /api/costs/by-agent

Response

GET /api/costs/by-model

Response

GET /api/costs/daily

Response

POST /api/costs/budgets

Request Body

GET /api/costs/budgets

DELETE /api/costs/budgets/{id}

Agent Risk Score Starter+ NEW

GET /api/risk-score?agent_name=X

Response

GET /api/risk-score/all

SDK Usage

Cost Prediction Starter+ NEW

GET /api/costs/predict?agent_name=X&model_used=Y

Response