Skip to content
WORKSI AI Control

MCP, with governance built in.

Agents report their work, check policies and ask for approval through the Worksi MCP server. Put any MCP server behind a gateway that filters tools and traces every call.

MCP client
{  "mcpServers": {    "worksi": {      "command": "npx",      "args": [        "-y",        "mcp-remote",        "https://gateway.worksi.ai/mcp",        "--header",        "Authorization:${WORKSI_AUTH}"      ],      "env": {        "WORKSI_AUTH": "Bearer aiw_…"      }    }  }}

One endpoint, two tool sets.

Connect to https://gateway.worksi.ai/mcp over Streamable HTTP. An API key gives an agent its agent tools; a personal access token gives a person the operator tools, with their role’s permissions.

Operator tools that change something are written to the audit log. Both credentials can read the worksi://docs/quickstart and worksi://status resources.

Agent toolsAPI key

Agent tools of the Worksi MCP server
ToolDoes
start_runStarts a run and returns its run and trace ids
log_stepRecords a step with status, duration, input and output
log_tool_callRecords a tool call with arguments, result and data source
end_runFinishes the run with its status
record_outcomeRecords a business outcome for ROI
check_toolIs this tool allowed, or does it need approval?
request_approvalAsks a person to approve something
get_approvalPolls an approval decision
get_statusIs the agent paused or the run stopped?

Operator toolspersonal access token

Operator tools of the Worksi MCP server
ToolDoes
list_incidentsOpen incidents, by status and severity
get_incidentEvidence headline, failure location and latest diagnosis
acknowledge_incidentAcknowledges an incident, with a note
list_agentsAgents and their status
pause_agentPauses an agent, under the console’s approval rules
resume_agentResumes a paused agent
list_approvalsPending approvals
decide_approvalApproves or rejects, with a note
get_usageYour plan and this month’s usage

Put any MCP server behind Worksi.

Register an upstream MCP server, its URL and an optional auth header stored encrypted, and agents reach it through Worksi with their API key.

tools/list
Shows only the tools your policies and the agent’s allowed tools permit.
tools/call
Checked first. A disallowed tool gets -32001; a tool that needs a person gets -32002 and a pending approval.
Every call
Recorded as a span. New tools appear in your AI inventory.
Pass-through
Resources, prompts, pings and notifications, unchanged.
Upstream URL
Checked against private‑network and cloud‑metadata addresses before every connection.
Clients connect to https://gateway.worksi.ai/mcp/servers/{slug} with an API key. An MCP client sends tools/call to the Worksi MCP gateway, which filters tools, enforces approvals and traces every call. Only allowed calls reach your MCP server.
.cursor/mcp.json
{  "mcpServers": {    "github-via-worksi": {      "url": "https://gateway.worksi.ai/mcp/servers/github",      "headers": {        "Authorization": "Bearer aiw_sk_live_…",        "X-AIW-Agent": "release-agent"      }    }  }}

Upstream servers per plan.

The Worksi MCP server itself is included in every plan. Plans differ in how many servers you can put behind the gateway.

Upstream MCP servers per plan
PlanUpstream MCP servers
Starter1
Pro and the trial10
BusinessUnlimited
EnterpriseUnlimited

Add Worksi to your MCP client.

Paste a configuration from the top of this page and restart the client. One request shows what a credential unlocks: agent tools for an API key, operator tools for a personal access token.

  • 3‑day trial with Pro features, no card.

Keep readingGovernanceAI agentsSDKs

Terminal
# the tools your API key or access token can callcurl -s https://gateway.worksi.ai/mcp \  -H "Authorization: Bearer $AIW_API_KEY" \  -H "Content-Type: application/json" \  -H "Accept: application/json, text/event-stream" \  -d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}'