এমসিপিModel Context Protocol, built into every instance
Build and govern @Human by talking to your AI assistant.
Every @Human instance runs its own MCP server. Connect any MCP-capable assistant and it can explore your projects, build data models, pages and automations, and write policies and risk scenarios, always as you, always within your permissions, and never past a human approval gate.
How it works: your assistant, your instance, your rules
MCP, the Model Context Protocol, is an open standard that lets AI assistants use tools. @Human exposes its platform as MCP tools, so an assistant can do what you could do in the console, through conversation.
Your AI assistant
Any MCP-capable client. You choose the assistant and the model behind it.
- Desktop and web assistants
- Code editors
- Your own agents
@Human instance · /api/mcp
Acts as the signed-in user. Same roles, same field-level permissions, same audit trail.
The whole platform, as tools
These are real tool names from a live @Human MCP server. They cover the full lifecycle, from the first module to the last compliance record.
Build
Data, pages, charts and apps
compose_namespace_createcompose_module_createcompose_page_createcompose_page_layout_updatecompose_chart_createcompose_record_reportsystem_application_source_set
Automate
Workflows, triggers and TAQs
automation_workflow_createautomation_trigger_createautomation_taq_createautomation_taq_script_testautomation_workflow_executionsautomation_taq_execution_trace
AI
Agents, chatbots and knowledge
system_agent_createsystem_agent_execsystem_chatbot_createsystem_llm_provider_lookupsystem_project_knowledge_base_create
Govern
Policies, risk and data
system_project_policy_createsystem_project_risk_framework_createsystem_project_risk_scenario_createsystem_project_sensitive_data_createsystem_project_dpia_createsystem_project_ai_system_create
Control
Guardrails, monitoring and access
system_project_runtime_enforcement_createsystem_project_monitoring_rule_createsystem_project_gate_lookupsystem_project_member_role_updatesystem_project_permission_allow
Manage and comply
Requests, notices and status
system_project_request_createsystem_project_end_user_notice_updatesystem_project_end_user_lookupsystem_project_statusdiscovery_search
human_tool_search so an assistant can find the right tool and read its documentation before using it.compose_record_draft puts a proposed record in front of you as an editable form without saving it, and approval gates can be looked up (system_project_gate_lookup) but never signed by a tool.Not a mock-up: real calls, real answers
These are tool calls made over MCP against a live @Human instance while we built this website, with the responses the instance returned.
Which projects do we have?
system_project_lookup{ "headsOnly": true }38 projects · 1 active ("Big Project") · 37 draftsThe assistant spotted two projects both named "LifeBridge", a likely duplicate.
What stops the support assistant promising refunds?
system_project_runtime_enforcement_lookup{ "title": "No refund or legal promises", "action": "block", "terms": 3, "fires": 2 }
{ "title": "No compensation wording", "action": "rephrase", "terms": 1, "fires": 1 }Two guardrails, linked to a high-severity risk scenario."No refund or legal promises" has blocked two replies.
Who must sign before this can go live?
system_project_gate_lookup→ quorum: all · evidence: required · lapse on change: yes → comply gate: required · 3 signatories
Every signatory must sign, with evidence; sign-off lapses on change.The comply gate needs the executive authority, governance owner and DPO.
Which rights does our framework assess?
system_project_risk_framework_lookup{ "name": "Fundamental rights (EU Charter)", "builtIn": true, "values": 41 }
{ "name": "Bangladesh environmental and biodiversity law", "values": 11 }41 rights in 7 chapters, plus a custom national framework.Custom frameworks sit side by side with the compulsory FRIA.
Responses trimmed for length. JSON lines are as returned by the instance; lines starting with → are summaries of longer responses.
Conversational, not uncontrolled
An assistant connected to @Human follows the same discipline as everyone else: propose, confirm, write, and only people approve.
- Acts as youThe assistant has exactly your roles and field-level permissions, nothing more
- Writes go to a draftLive versions change only after signed approval gates
- Every change is reviewedNew data, purposes or AI components raise an impact review automatically
- Everything is loggedChanges and AI activity are recorded in the event log
This website was researched over MCP
The data behind the screens on our site was read from a live @Human instance through its MCP server: project structures, records, chart definitions, guardrails and frameworks. Here's some of it.
Connect in three steps
Works with any assistant or tool that supports MCP. Nothing to install on the @Human side: the server is already running on your instance.
- Copy your instance's MCP addressEvery instance serves MCP at
/api/mcp:https://YOUR-INSTANCE/api/mcp - Add it to your assistantAdd it as a custom connector or MCP server, and sign in with your own @Human account when asked.
- Start askingExplore projects, build, and draft governance records. Writes are proposed for your confirmation.
Try asking
Create a module for site reports with title, location, severity and status, then a list page for it.
Which risk scenarios in this project have no runtime enforcement or monitoring rule?
List every place personal data lands in this project, including prompts and agent memory.
Summarise what's waiting for my signature, and what changed since the last approval.
Which AI assistants can I use?
Any MCP-capable client, including desktop and web assistants, code editors and your own agents. @Human doesn't lock you into one assistant or model.
What can the assistant access?
Exactly what your @Human user can access, with the same roles and field-level permissions. It cannot see more than you can.
Can an assistant publish or approve changes?
No. Writes land in a draft version and trigger an impact review. There is no MCP tool to sign an approval gate or publish a version; that stays with named people.
Where does my data go?
@Human runs in your own infrastructure. Results the assistant reads are processed by the AI provider behind that assistant, so choose a provider that fits your data rules, or keep sensitive projects out of scope.
Do I need AI to use @Human?
No. Everything available over MCP is also available in the @Human console, and your applications keep running without any AI connected.