Forge-mcp is a cross-manufacturer industrial MCP server developed by Foundry Labs. It is designed to solve the problems of industrial equipment data silos and difficult AI agent access, connecting various industrial machines (such as CNCs, robots, and industrial equipment) to AI agents through standardized interfaces.
-
Main Purpose The main purpose of Forge-mcp is to provide a unified bridge that allows AI agents to interact with any CNC, robot, or industrial machine in natural language. It focuses on machine identity management, telemetry data standardization, automation, and work records.
-
Main Capabilities
-
Machine identity and data standardization: Provides tools such as identify_machine, normalize_telemetry, query_machine_history, get_coverage, and correct_mapping. It can normalize raw OEM telemetry data from 18 manufacturer families into a unified canonical vocabulary, containing thousands of verified field mappings, so agents can write logic against the same set of field names without caring whether the underlying system is a Fanuc CNC, a KUKA robotic arm, or a Universal Robots collaborative robot.
-
Natural language automation: Provides tools such as create_automation, activate_automation, list_automations, disable_automation, delete_automation, restore_automation, and query_webhook_history. It can parse plain English instructions into structured automation workflows, which users can review before activating.
-
Prediction and diagnostics: Based on the TimesFM model, provides tools such as predict, predict_breach, remaining_life, predict_batch, fleet_health, detect_anomalies, machine_intelligence, and prediction_accuracy. These tools are used to predict failures, assess remaining life, score fleet health, and detect anomalies.
-
Operations management: Provides tools such as calculate_oee, fleet_oee, energy_consumption, shift_report, diagnose_machine, and health_index, used to calculate OEE, energy consumption, generate shift reports, and health indices.
-
Guardrails and records: Provides guardrail tools such as check_guardrail and list_guardrails, as well as verify_record for creating tamper-proof, hash-verified work records.
-
Agent coordination: Provides get_agent_card and list_agents, used to manage agents connected to your kernel instance (their capabilities, trust scores, and machine access permissions). These tools are included in the subscription, with no per-call charges.
-
Demonstration and testing: Provides fire_sandbox for demonstrating the complete watch → fire → settle loop.
-
How to Use Forge-mcp is a pure HTTP proxy, and each tool is a wrapper for https://forge.foundrynet.io/v1/* . It is hosted on Streamable HTTP, with the MCP endpoint at https://mcp.foundrynet.io/mcp，同时保留旧版 SSE endpoint https://mcp.foundrynet.io/sse。 Connection methods support Claude Desktop, Cursor, or any MCP client. It can be configured via claude_desktop_config.json together with the mcp-remote bridge. Users need to obtain a free fnet_ key, available at https://foundrynet.io/signup Registration (provides 50 normalized calls, no credit card required). The server side requires environment variables, and the codebase includes files such as mcp_server.py, gating.py, server.json, smithery.yaml, requirements.txt, etc. Deployment is carried out via the deploy.sh script, which performs release gating, deployment locking, build stamping, and health check verification.
IV. Applicable Scenarios Forge-mcp is suitable for scenarios that require connecting AI agents to industrial equipment, such as smart manufacturing, the industrial Internet of Things, predictive equipment maintenance, cross-brand device data integration, automated workflow orchestration, and compliance scenarios that require tamper-proof work records.
V. Limitations
- The free tier provides only read-only tools; advanced tools such as prediction and diagnostics require pay-per-use.
- An API key must be obtained and configured.
- Deployment and production use must follow a specific release process.
- Some advanced features (such as cross-platform agent trust) require reference to the Assay platform.
- The server side requires correct environment configuration and dependencies.
- Tool calls require a valid tenant context, and OAuth access tokens may not be accepted.
