Mitti MCP Server
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Mitti MCP Server show me my open actions"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Mitti MCP Server
A production-grade Model Context Protocol (MCP) server for Mitti (the 2026 rebrand of SafetyCulture — see the rebrand FAQ), built with FastMCP, Pydantic v2, and async httpx.
Expose your Mitti inspections, templates, actions, and users as MCP tools for any LLM client (Claude Desktop, Claude Code, Cursor, Gemini CLI, and more).
Migrated from SafetyCulture to Mitti.
api.safetyculture.iostill works with no announced shutoff date, butapi.mitti.comis the long-term host and is now the default. LegacySAFETYCULTURE_API_TOKEN/SAFETYCULTURE_BASE_URLenv vars still work as a fallback. See API limitations below — several endpoints changed shape, not just host, in the rebrand.
Features
Domain | Tools Available |
Inspections | List (lightweight), get (full detail), complete, web report link, visual table view |
Templates | List (lightweight), get (full detail), get definition |
Actions | List, get, create, update status, update title |
Users | List, search (client-side filtered) |
Related MCP server: Sentinel Core Agent
API limitations (read this before relying on a field)
The Mitti API is not a drop-in rename of the old SafetyCulture API — verified against developer.mitti.com/reference:
list_inspectionsonly returnsaudit_id,template_id, anddate_modified— the search endpoint'sfieldparameter doesn't support name/owner/score/status. Callget_inspectionwith a specificaudit_idfor full detail.list_templatessimilarly only returnstemplate_id,name, and dates. Callget_templatefor description/owner/archived status.Action priority is now an org-specific
priority_id(UUID from your org's task-type configuration), not a fixedlow/medium/highstring.create_actionaccepts an optionalpriority_id— omit it for the system default.Action status IS still a fixed, documented set for the built-in Actions type, so
update_action_statusandlist_actions'sstatusfilter keep the friendlyopen/in_progress/completed/cant_dointerface — translated under the hood to Mitti's stable status UUIDs.list_actionsfilters status andpriority_idserver-side (via the API'stask_filters), not by fetching everything and filtering in Python.get_current_useris unsupported. Mitti's current API has no "current authenticated user" / "me" endpoint at all. The tool returns a clear failure explaining this — usesearch_userswith a known email instead.User status and role are proto enums on the wire (e.g.
USER_ACTIVE_STATUS_ACTIVE,SUBSCRIPTION_SEAT_TYPE_PREMIUM) —list_users'sstatusfilter accepts the friendlyactive/inactive(there is nopendingstatus), andUserSummary.status/.roleare translated back to short lowercase labels rather than leaking the raw wire values.search_usersdoes client-side substring filtering, because the replacement for the deprecated email-only search endpoint (/users/v1/users/list) only supports exact-match filters, not free text.
Requirements
Python 3.12+
uv (package manager)
A Mitti API token (see the Mitti Developer Portal)
Setup
1. Clone & install
git clone <your-repo>
cd mitti-mcp
uv venv --python 3.12
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate
uv add "fastmcp[apps]" httpx pydantic python-dotenv2. Configure your API token
cp .env.example .env
# Edit .env and set MITTI_API_TOKEN=your_token_hereTo generate a Mitti API token:
Log into Mitti → Account Settings → API Tokens
Click Generate Token and copy the value.
3. Run and test the server locally
Option A: Interactive Inspector UI (Development Mode)
FastMCP includes a browser-based developer inspector interface. Run the command:
fastmcp dev inspector src/mitti_mcp/server.pyThis starts your server and prints a link containing an authentication token (e.g., http://localhost:6274/?MCP_PROXY_AUTH_TOKEN=...). Open this link in your browser to test each tool.
Alternatively, you can run the inspector without authentication:
# Windows PowerShell
$env:DANGEROUSLY_OMIT_AUTH="true"; fastmcp dev inspector src/mitti_mcp/server.py
# macOS/Linux/Git Bash
DANGEROUSLY_OMIT_AUTH=true fastmcp dev inspector src/mitti_mcp/server.pyOption B: Inspect definitions via CLI
Check the tool schemas and registered components directly on the command line:
fastmcp inspect src/mitti_mcp/server.pyOption C: Call tools from the CLI
You can query and execute specific tools directly from the terminal:
# List all tools and parameters
fastmcp list src/mitti_mcp/server.py
# Execute a tool with parameters
fastmcp call src/mitti_mcp/server.py list_users limit=5Option D: Run directly via Python
To run the server directly (using stdio transport):
uv run python -m mitti_mcp.serverClaude Desktop Configuration
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"mitti": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/mitti-mcp",
"run",
"python",
"-m",
"mitti_mcp.server"
],
"env": {
"MITTI_API_TOKEN": "your_token_here"
}
}
}
}Connecting to the remote (Render) deployment via HTTP
For a client that supports remote "type": "http" MCP servers directly (no stdio bridge
needed), point it at the deployed URL and pass your Mitti API token as a custom header —
not Authorization, since FastMCP's HTTP layer reserves that header for its own auth
and strips it before it reaches tool code:
{
"mcpServers": {
"mitti": {
"type": "http",
"url": "https://mitti-mcp.onrender.com/mcp",
"headers": {
"x-mitti-token": "your_token_here"
}
}
}
}client.py checks the incoming request's x-mitti-token header first, falling back to
the server's own MITTI_API_TOKEN env var if the header is absent. This means:
One shared deployment can serve multiple callers/orgs, each with their own token, with no server-wide token configured on Render at all.
Or, keep
MITTI_API_TOKENset in Render's dashboard as before, and the header becomes optional — useful while testing, since you don't have to configure it in every client.
To make the header mandatory (no shared fallback), remove MITTI_API_TOKEN from
Render's environment variables — then any request without a valid x-mitti-token header
fails clearly instead of silently using a shared token.
Deployment (Prefect Horizon)
To deploy your server live to the cloud:
Push your repository to GitHub (public or private).
Sign in to horizon.prefect.io.
Connect your GitHub account and select your repository.
Configure your deployment:
Server name: A unique name for your server (this determines your server's endpoint URL).
Entrypoint:
src/mitti_mcp/server.py:mcpAuthentication: Enable to secure your server endpoints with OAuth.
Click Deploy Server to build and launch your live production endpoint.
Deployment (Render)
A Render Blueprint is included (render.yaml)
that runs the server over streamable-HTTP instead of stdio:
Push this repo to GitHub.
On render.com, click New + → Blueprint and select this repo.
Render reads
render.yamlautomatically. When prompted, set theMITTI_API_TOKENenvironment variable (it's intentionally left blank in the blueprint — never commit a real token).Deploy. The MCP endpoint is served at
https://<your-service>.onrender.com/mcp(not/) — point streamable-HTTP MCP clients at that path.
Not yet verified against a live Render deploy: Render's default health check hits /,
which this server doesn't serve (only /mcp). If the service is marked unhealthy
despite running fine, set healthCheckPath: /mcp in render.yaml or add a trivial
health route.
Project Structure
mitti-mcp/
├── README.md
├── pyproject.toml
├── render.yaml
├── .env.example
└── src/
└── mitti_mcp/
├── __init__.py
├── server.py # FastMCP app entry point
├── client.py # httpx Mitti API client
├── tools/
│ ├── inspections.py # Inspection tools
│ ├── actions.py # Action tools
│ ├── templates.py # Template tools
│ └── users.py # User tools
└── models/
└── schemas.py # Pydantic v2 modelsTech Stack
Layer | Tool |
MCP framework | FastMCP 4.x |
HTTP client | httpx (async) |
Validation | Pydantic v2 |
Python | 3.12+ |
Package manager | uv |
Middleware
The server applies cross-cutting middleware (src/mitti_mcp/server.py) to every
tool call across all mounted sub-servers, in this order:
ErrorHandlingMiddleware— catches and logs unhandled exceptions consistently.LoggingMiddleware— logs every request/response with duration and payload size.RateLimitingMiddleware— global token-bucket limit (default 10 req/s, override withMITTI_MCP_RATE_LIMIT_RPS) to protect the Mitti API from bursts.
Human-in-the-Loop Approval (Prefab UI)
Requires fastmcp[apps] (already in pyproject.toml). The server registers FastMCP's
Approval provider, which exposes a request_approval tool. The server instructions
and the descriptions of all write tools (complete_inspection, create_action,
update_action_status, update_action_title) tell the LLM to call request_approval
and wait for the user's decision before performing the write.
The user sees an Approve/Reject card rendered inline in the conversation (an MCP Apps
UI component, not plain text). Clicking a button sends "<summary>" — I selected: Approve/Reject back into the conversation as if the user typed it.
This is an advisory gate, not server-side enforcement — it relies on the LLM
following the instructions. It does not replace the input validation and error
handling already in client.py and the tool modules.
Try it: fastmcp call src/mitti_mcp/server.py request_approval summary="..."
Visual inspection browsing
list_inspections_view (app=True) returns the same data as list_inspections but as
an interactive, sortable, searchable DataTable rendered inline in the conversation
instead of raw JSON. The LLM picks this tool over list_inspections when the user
wants to browse/sort/search visually.
License
MIT
Available Tools
17 toolsactions_create_actionActions Create ActionA
Create a new action in Mitti. Requires at minimum a title. Optionally set description, due date (ISO 8601), assignee user IDs, and site ID. priority_id is an org-specific UUID — omit it for the system default priority. Call request_approval first and wait for the user's decision before calling this tool. Returns the created action's ID.
Inputs: title (required), due_at, site_id, description, +2 more. Returns: Response after successfully creating an action.
actions_get_actionActions Get ActionA
Get full details of a specific Mitti action by its ID. Returns title, description, status, priority_id, due date, assignees, and site.
Inputs: action_id (required). Returns: Condensed action record.
actions_list_actionsActions List ActionsA
List actions from the Mitti (formerly SafetyCulture) account. Supports server-side filtering by status (open, in_progress, completed, cant_do) and by an org-specific priority_id (UUID). Returns action ID, title, description, status, priority_id, due date, and assignees.
Inputs: limit, status, priority_id. Returns: Result of listing actions.
actions_update_action_statusActions Update Action StatusA
Update the status of a Mitti action. Valid statuses are: 'open', 'in_progress', 'completed', 'cant_do'. Call request_approval first and wait for the user's decision before calling this tool. Returns success or an error message.
Inputs: status (required), action_id (required). Returns: Generic success/failure result for write operations.
actions_update_action_titleActions Update Action TitleA
Update the title of a Mitti action. Call request_approval first and wait for the user's decision before calling this tool. Returns success or an error message.
Inputs: title (required), action_id (required). Returns: Generic success/failure result for write operations.
inspections_complete_inspectionInspections Complete InspectionA
Mark a Mitti inspection as complete. This is a write operation — the inspection must be in 'in_progress' status. Call request_approval first and wait for the user's decision before calling this tool. Returns success or an error message.
Inputs: audit_id (required). Returns: Generic success/failure result for write operations.
inspections_get_inspectionInspections Get InspectionA
Get full details of a single Mitti inspection by its audit ID. Returns name, template, owner, dates, score, and status.
Inputs: audit_id (required). Returns: Full inspection record with metadata and item summary.
inspections_get_inspection_web_report_linkInspections Get Inspection Web Report LinkA
Get a shareable web report link for a Mitti inspection. Returns a permanent URL that can be shared with others to view the inspection report.
Inputs: audit_id (required). Returns: Generic success/failure result for write operations.
inspections_list_inspectionsInspections List InspectionsA
List inspections from the Mitti (formerly SafetyCulture) account. Supports filtering by modification date and limiting the number of results. Returns only audit_id, template_id, and modification date per inspection — the search API does not return name, owner, score, or status. Call get_inspection with a specific audit_id for full details.
Inputs: limit, template_id, modified_after. Returns: Result of listing/searching inspections.
inspections_list_inspections_viewInspections List Inspections ViewA
Show inspections from the Mitti account as an interactive, sortable, searchable table rendered directly in the conversation. Use this instead of list_inspections when the user wants to browse inspections visually. Only shows audit_id, template_id, and modification date — the same limitation as list_inspections.
Inputs: limit, template_id, modified_after.
request_approvalRequest ApprovalA
Request human approval before proceeding with an action.
Call this tool proactively whenever you are about to take a significant or irreversible action and want the user to confirm first. Do NOT wait for the user to ask you to seek approval — use your judgment about when confirmation is appropriate.
The user will see an approval card with the summary, optional details, and Approve/Reject buttons. When they click a button, their decision appears as a message in the conversation (as if the user typed it), like:
"Deploy v3.2 to production" — I selected: Approveor:
"Deploy v3.2 to production" — I selected: RejectIMPORTANT: After calling this tool, you MUST stop and wait for the user's response. Do not continue, do not take any other actions, do not generate further output until you see the "I selected:" message. If approved, continue with the action. If rejected, acknowledge and ask how to proceed.
Inputs: summary (required), title, details, reject_text, +3 more.
templates_get_templateTemplates Get TemplateA
Get summary metadata for a specific Mitti template by its ID. Returns name, description, owner, archived status, and dates.
Inputs: template_id (required). Returns: Condensed template record.
templates_get_template_definitionTemplates Get Template DefinitionC
Get the full definition of a Mitti template including all questions, sections, and scoring rules. Returns a structured summary of the template definition.
Inputs: template_id (required). Returns: Generic success/failure result for write operations.
templates_list_templatesTemplates List TemplatesA
List templates available in the Mitti (formerly SafetyCulture) account. Returns template ID, name, and dates only — description, owner, and archived status require get_template. Use the returned template_id to get the full template definition.
Inputs: limit, archived. Returns: Result of listing templates.
users_get_current_userUsers Get Current UserA
Get the profile of the currently authenticated Mitti user. NOTE: Mitti's current API has no dedicated 'current user' endpoint — this always returns failure with guidance to use search_users with a known email instead.
Inputs: no named parameters. Returns: Generic success/failure result for write operations.
users_list_usersUsers List UsersA
List users in the Mitti (formerly SafetyCulture) organization. Returns user ID, first name, last name, email, status ('active' or 'inactive' — there is no 'pending' status), and a best-effort role indicator. Useful for finding user IDs to assign to actions.
Inputs: limit, status. Returns: Result of listing users.
users_search_usersUsers Search UsersA
Search for Mitti users by name or email substring. NOTE: the underlying API only supports exact-match filters, so this fetches all users and filters client-side — it is not efficient for very large organizations. Returns matching users with their IDs, names, emails, and statuses.
Inputs: query (required), limit. Returns: Result of listing users.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 17 tools
Most tools are clearly separated by resource (templates, inspections, actions, users) and action (list, get, create, update). The main ambiguity is between inspections_list_inspections and inspections_list_inspections_view, which return the same data but differ only in presentation format, and templates_get_template_definition vs templates_get_template could confuse agents looking for template details.
Tool names follow a consistent resource_action pattern (e.g., inspections_list_inspections, actions_get_action, users_search_users). Minor deviations include request_approval lacking a resource prefix and templates_get_template_definition being more verbose than the pattern, but overall the convention is predictable.
17 tools is slightly above the ideal range but reasonable for a server covering four distinct resource domains (templates, inspections, actions, users) plus an approval utility. Each tool serves a clear purpose, though a few could be consolidated (e.g., list_inspections vs list_inspections_view).
The server covers the main workflows: listing and retrieving templates, inspections, actions, and users, plus creating actions and updating action status/title. Gaps include no create/update for templates or inspections, no delete operations, and no way to update inspection details, but the core read and action-management workflows are covered.
Maintenance
Related MCP Connectors
Find, vet, and run MCP tools through a secure audited gateway with prompt-injection risk scoring
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Your org's AI agents, tasks, runs, search, and brain files as MCP tools and resources.
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables LLM agents to programmatically inspect, debug, and test other MCP servers by wrapping the MCP Inspector CLI. Supports listing and calling tools, reading resources, and testing prompts on both local and remote MCP servers.65 npmMIT
- FlicenseNot gradedqualityDmaintenanceEnables file system operations, web scraping, and AI-powered search through MCP tools for use by LLM agents.1-
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to discover and execute tools via a secure MCP server with JWT authentication, RBAC, rate limiting, and audit logging.1MIT
- FlicenseNot gradedqualityDmaintenanceEnables LLM-powered agents to securely communicate with and orchestrate downstream microservices via FastAPI endpoints exposed as MCP tools.-