Agent integration

Structured tools for agents that need action-ready Planetary Model context.

AgentLayer gives approved agents a governed way to search entities, read context, fetch evidence, inspect decision inputs, follow updates, and run bounded scenarios.

Tool categories

Enough structure for agents. Enough restraint for institutions.

world.read

Entity discovery

Find canonical entities, private watchlists, and package scope before requesting records.

world.read

Gap discovery

Search for stale assumptions, missing evidence, hidden exposure, and fixable operational gaps inside approved Planetary Model context.

world.read

Planetary Model reads

Read current entity records, recent changes, grounded context packets, composite scores, domain packages, coverage, blockers, and next-action posture.

evidence.read

Evidence reads

Fetch evidence bundles and source references before citing a claim.

decision.read

Decision inputs

Read typed decision records for governed workflows.

stream.read

Subscribed updates

Read entitled updates for subscribed records and packages.

simulate.run

Scenarios

Run bounded scenario analysis with explicit access.

webhook.manage

Webhook visibility

Read entitled webhook endpoint and delivery records under read-only visibility. External webhook creation and customer secret rotation are not currently approved.

Read-first

Agents query context, evidence, packages, and updates before considering any downstream workflow.

Scoped

Sensitive capabilities require explicit authorization and may return redacted outputs.

Inspectable

Responses preserve enough context for a human, system, or auditor to understand what was used.

Bounded

Scenarios are separate from observed records and require explicit access.

Connection shape

AgentLayer is meant to be connected through approved access.

Approved clients receive the current source or package distribution instructions and scopes needed for their use case. The public package command is not published yet; AgentLayer MCP uses the governed AgentLayer module and the same approved bearer token policy as other AgentLayer integrations.

Example MCP config

{
  "mcpServers": {
    "crowdalpha-agentlayer": {
      "command": "python",
      "args": ["-m", "services.agentlayer.mcp_server"],
      "env": {
        "CROWDALPHA_API_BASE": "https://api.crowdalpha.ai",
        "CROWDALPHA_API_TOKEN": "$CROWDALPHA_API_TOKEN"
      }
    }
  }
}

Non-runtime verification

# approved source/package access
export CROWDALPHA_API_BASE="https://api.crowdalpha.ai"
export CROWDALPHA_API_TOKEN="ca_agent_..."

python -m py_compile services/agentlayer/mcp_server.py
python -m pytest services/agentlayer/test_mcp_contract.py -q
pnpm agent:discovery-check

Agent integrations should carry context, evidence, and provenance with every serious response.

Request AgentLayer accessView developer contracts