Install the APIANT Claude Code Plugin and Claude Code gains the ability to build integrations, edit assemblies, run tests, deploy to production, write docs, and diagnose customer issues. Below is the complete inventory: 42 skills, 138 MCP tools, each one a real capability Claude can invoke on your behalf.
The APIANT Claude Code plugin ships three MCP servers and a skills library. Install it, open Claude Code, and the entire platform becomes addressable through natural language.
Run once. The plugin verifies MCP connections, env vars, and plugin version. Claude announces what's live and what's missing.
"Build a Mindbody to Shopify sync." Claude Code picks the right skill, activates the right toolset, and starts executing with live feedback.
Claude composes assemblies, tests end-to-end, fixes failures, deploys to prod, and monitors the account, without handoffs.
Every skill is a procedure a senior APIANT engineer follows, encoded and executable by Claude Code. Click any card for the full what/when/how.
Walk the user through installing or repairing the APIANT Claude Code plugin via the `apiant-cli` Node CLI. Use when the user says "setup", "install", "reinstall", "update plugin", or on first run.
Build a new APIANT automation using MCP tools. Use when the user says "build an automation", "create an automation", "connect X to Y", or describes a new workflow.
Build APIANT app assemblies (API integration plumbing) using MCP tools. Use when the user says "build an assembly", "create a connector", "add an API action", or describes API integration work.
Build a multi-automation integration suite on APIANT. Use when the user says "build an integration", "create a two-way sync", "build a data pipeline", or any request involving multiple coordinated automations.
Design a form for an APIANT automation. Use when the user says "build a form", "design a form", "create a form", or describes a form they need.
Edit an existing APIANT automation using MCP tools. Use when the user says "edit automation", "add a step", "fix the mapping", "swap X for Y", "move step", "disable step", or describes changes to an existing automation.
Edit an existing APIANT assembly using MCP tools. Use when the user says "edit assembly", "fix the JSP", "update the API call", "change the endpoint", "rename a setting", "edit the dropdown", or describes changes to an existing assembly's code or settings.
Test and validate an APIANT automation end-to-end. Executes it, inspects results, diagnoses failures, applies fixes, retests, and runs branch coverage. Use when the user says "test this automation", "run a test", "why is automation X failing", or "diagnose automation X".
Integration test a multi-automation suite on APIANT. Verifies data flows between automations, shared state consistency, loop prevention, and cross-automation coordination. Use after building an integration suite or when the user says "test the integration end to end".
Publish template automations from dev to prod and deploy them to customer accounts (linked or unlinked, user's choice per deploy). Use when the user says "publish", "deploy", "push to prod", or "update customers".
Type-specific instructions for building a NEW_ITEM polling trigger assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building a NEW_OR_UPDATED polling trigger assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building a WEBHOOK (manual registration) trigger assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building a WEBHOOK_SELF_REGISTERING trigger assembly and its companion Delete webhook action. Invoked by /build-assembly after type classification.
Type-specific instructions for building a SERVICE_WEBHOOK trigger assembly (credential-based webhook URL with event filtering). Invoked by /build-assembly after type classification.
Type-specific instructions for building protocol thread triggers (PROTOCOL_THREAD_RECEIVER and PROTOCOL_THREAD). Invoked by /build-assembly after type classification.
Type-specific instructions for building an ADD action assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building a DELETE action assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building a FIND action assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building a GET action assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building a LIST action assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building an UPDATE action assembly. Invoked by /build-assembly after type classification.
Type-specific instructions for building an INVOKE action assembly: RPC-style POST that returns rich data (AI inference, generation, computation, transformation). Invoked by /build-assembly after type classification.
Type-specific instructions for building an app connector assembly (OAuth V2, API Key, or No Credentials). Invoked by /build-assembly during Phase 2.
Register an OAuth app on a vendor's developer portal by driving a browser, capturing client_id and client_secret, and storing them in the APIANT keyvault. Use when the user says "register oauth app", "get oauth credentials", "sign up for oauth", or asks to automate getting client ID/secret from a vendor like Salesforce, HubSpot, Google, etc.
Convert a human-built APIANT assembly to the AI template pattern. Use when asm_load returns is_human_built=true, or the user says 'convert assembly', 'migrate assembly', 'convert to AI template'.
Type-specific instructions for building two-way sync (bidirectional) trigger and action assemblies. Invoked by /build-assembly after type classification.
Pattern reference for chat widget automations, from simple AI chatbots to multi-goal conversation flows with tool automations. Loaded by /build-automation when the intent involves chat widget, chatbot, chat goal, conversation memory, AI agent with tools, or multi-turn chat.
Pattern reference for importing field mappings from a CSV or spreadsheet. Loaded by /build-automation when the user provides a CSV file defining how fields should map between systems.
Pattern reference for Execute Automation chaining, parent/child automations, query string parameter passing, webhook payload forwarding. Loaded by /build-automation when the intent involves executing, calling, or chaining automations.
Pattern reference for human moderation, pause an automation until a person approves or denies an item via a moderation queue link. Loaded by /build-automation when the intent involves human approval, review, moderation, decision gates, or approve/deny workflows.
Pattern reference for fan-out/fan-in automation latches, launch N child automations in parallel, wait for all to finish. Loaded by /build-automation when the intent involves parallel execution, fan-out, fan-in, or latch groups.
Pattern reference for snoozing/pausing automations until a future datetime. Loaded by /build-automation when the intent involves snoozing, delaying, pausing, sleeping, or waiting until a specific time.
Pattern reference for Collector automations, aggregate items into named buckets across executions, then drain the bucket on a schedule or trigger. Loaded by /build-automation when intent involves "aggregate", "collect over time", "daily digest", "batch", "bucket", or "send a summary email of today's items".
Pattern reference for MCP tool automations, automations exposed to MCP clients (Claude Desktop, Cursor, ChatGPT MCP, etc.) as callable tools. Loaded by /build-automation when the intent involves "expose to Claude", "expose to MCP client", "MCP tool", "Model Context Protocol", or "make this callable from an AI assistant".
Cookbook of small system-utility recipes, Feed (RSS/Atom/RDF), PDF generation, Datafile (daily CSV), and other one-action utilities. Loaded by /build-automation when intent matches "RSS", "Atom", "feed", "publish a feed", "monitor a news source", "generate a PDF", "PDF attachment", "log to CSV", "datafile", "daily CSV export", or "feed data into ChatGPT".
Pattern reference for web service automations, synchronous HTTP endpoints where the caller waits for a response. Loaded by /build-automation when the intent involves "expose as an HTTP endpoint", "REST API", "synchronous webhook with response", "callable from another system", or "automation that returns data to the caller".
Monitor a customer account's automations for execution errors on prod. Use when the user says "monitor account", "check for errors", "watch this account", or after deploying automations to a customer. Designed for recurring use via /loop.
Diagnose customer automation issues, halt misbehaving runs, bulk-disable an account group as an incident kill-switch (with later restore), and provision new customer accounts. Searches execution history, traces failures, extracts HTTP requests, halts runaway automations, snapshots and disables all currently-on automations across a parent account plus its linked children, then restores them after the incident. Creates new APIANT accounts for integration products via admin_create_account. Use when support says "why did X not process", "check customer Y", "what's broken", "stop this automation", "turn off all automations for X", "kill switch", "disable across parent and children", "GHL outage", "the API is choking the queue", "restore automations after the incident", "what's still disabled from incident X", "create a new account for [product]", "provision an account", "spin up an account for [integration]".
Inspect, tune, or debug how APIANT handles automation errors, retry eligibility, 401 auto-turnoff, alert message transforms, and the dismissal list. Use when the user says "alert mapping", "suppress alert", "alert rule", "noisy alerts", "why did my error message say...", "my alert isn't firing", "why didn't my automation retry", "automation got turned off with 401", "add to retryable errors", "ignore 401 for this domain", or asks why an error email was rewritten.
Answer questions about APIANT grounded in the public documentation at info.apiant.com, how things work, how to do something, what a feature is, where to find a setting. Use when the user asks "how do I", "what is", "where is", "explain", "what does X do", or any platform question that isn't a build, test, or diagnose request.
File a bug report to APIANT's internal triage when you observe the MCP server itself returning wrong or broken behavior, wrong response shape, contract drift between docs and behavior, unhandled errors, schema violations, missing required fields in the response, regression on inputs that previously worked, or documented behavior that doesn't actually work. NOT for your own tool misuse, bad user input, or platform behavior that's working as designed. Autonomous, file the report immediately when you see it, don't ask the user for permission first. Use when you see "tool returned wrong shape", "schema violation", "this should work but doesn't", "contract drift", "MCP returned an unhandled error", "the tool description says X but it does Y", "the response is missing a required field", "this used to work", "documented behavior is broken", "regression in [tool]", or any time you're tempted to silently work around an MCP-layer bug instead of reporting it.
Skills drive the workflow. MCP tools do the actual work against APIANT's dev, prod, and docs environments. Tools are grouped into toolsets. Only the core set is always loaded; the rest activate when a skill asks for them, so the agent's context stays lean and the capability stays deep.
Load the specific toolset needed for the current skill. Keeps the context window lean.
Load, list, create, update, rename, and delete assemblies. The low-level assembly API.
Structured overview of an automation: triggers, steps, mappings. Fast context for edits.
Flat list of every automation in a scope. Filter by name, folder, or status.
Navigate the folder tree. Find automations grouped by customer, product, or environment.
Create, rename, duplicate, move, enable, disable. Every CRUD operation for automations.
The full catalog of available app integrations. Filter by category, vendor, or capability.
21 tools. Search the catalog, compose triggers and actions, edit mappings, commit versions, save patterns. The full automation authoring surface.
17 tools. Run test executions, inspect every step's input and output, mutate data, replay webhooks, walk branch-coverage test points. Debugging with full ground truth.
14 tools. List and organize committed automations, version them, compare versions, roll back. The git-like layer over automation drafts and releases.
24 tools. Switch between customer accounts, query usage and health across a tenant, read the keyvault, run ad-hoc SQL, tune retryable errors, manage lookups. The operator's toolbox.
6 tools. Tenant-wide alert rules, suppressed-alert lists, retry policy, 401 domain carve-outs. Requires Switch Account context.
10 tools. Per-automation alert mapping rules, step-level alert triggers, mapping trace, suppression, system-level mappings.
3 tools. Search, save, and delete reusable pattern recipes: field mappings, transforms, gotchas, templates.
7 tools. Publish folders or individual assemblies from dev to prod. Roll changes out to linked customer accounts in a single coordinated step.
35 tools. Create connectors, inject settings, compile JSP, test API endpoints, wire dynamic field discovery, finalize assemblies. The low-level integration plumbing.
Programmatic access. Issue a docs API key for CI/CD or automated publishing.
Retrieval-augmented Q&A. Claude asks the docs corpus before writing code or answering.
Publish a new doc. Title, body, parent, tags, and status, in one call.
Browse the full table of contents. Filter by section, tag, or status.
Authenticate the docs session. Tokens cached for the session.
Fetch the full contents of any doc. Markdown, images, and metadata included.
Take a live screenshot of the editor or platform UI and embed it into a doc.
Semantic + keyword search across the entire docs corpus. Ranked results with context.
Edit title, body, tags, or status. Revisions tracked automatically.
Attach an illustration or diagram to any page. Auto-alt-text generated.
List revisions, diff between versions, roll back if needed.
''Try a shorter query, remove a filter, or clear the search.
A SaaS team runs a boutique fitness CRM on HubSpot. Two thousand paying customers want their HubSpot contacts to sync two-way with Mindbody clients. Classic white-label integration. Historically: four to six weeks of engineering.
catalog_list_apps({ vertical: 'wellness' }) -> hubspot, mindbody auto_folder({ action: 'create', path: '/customers/boutique-fit' }) -> ok pattern_search({ shape: 'bidirectional-sync' }) -> 2 matches
auto_build({ trigger: 'assembly-trigger-updated', source: 'hubspot.contact' }) auto_edit_structure({ add_step: 'assembly-action-find', app: 'mindbody.client' }) auto_edit_structure({ add_step: 'assembly-action-add-or-update' }) auto_edit_mapping({ fields: [email, firstName, lastName, phone, membership_tier] })
HubSpot stores tier as a free-text dropdown. Mindbody expects one of four enum values. Claude detects the mismatch, adds a lookup-table transform, and saves it as a pattern so the reverse direction reuses it.
auto_build({ trigger: 'assembly-trigger-new', source: 'mindbody.client', interval: '5m' }) pattern_search({ name: 'membership_tier_enum' }) -> reusing from A auto_edit_mapping({ reverse: true, pattern: 'membership_tier_enum' })
Mindbody webhooks silently drop events during high-volume periods. Claude defaulted to polling with a 5-minute cursor, and noted this as a known-gotcha in the pattern library so future Mindbody integrations get the same treatment.
asm_set_action_throttle({ assembly: 'hubspot.contact.update', rate: 100, window: '10s' }) asm_set_connector_throttle({ connector: 'hubspot.v3', burst: 20 })
Throttling per automation would let any other HubSpot flow exhaust the budget. Throttling at the assembly (connector + action) applies the limit across every automation in the tenant, which is the only correct level for a shared API.
exec_test_automation({ direction: 'A', input: <synthetic> }) -> ok (1.2s) exec_test_automation({ direction: 'B', input: <synthetic> }) -> ok (1.8s) exec_get_assembly_log({ check: 'echo_loop' }) -> suppressed 14/14 test cases pass
deploy_publish_assembly({ folder: '/customers/boutique-fit' }) -> v1.0.0 deploy_list_linked_accounts({ template: 'hubspot-mindbody-sync' }) -> 2000 accounts deploy_to_accounts({ batch_size: 50 }) -> 2000/2000 ok
What the builder did not do: write a single line of JSP. Hand-map a single field. Open the Mindbody developer docs. File a ticket about the tier enum. Wait four to six weeks.
> Customer Alpine Clinic says nothing synced last night. ● /support searching execution history found 47 failures at 02:14 UTC root cause: Cliniko API returned 429 (rate limit) during bulk backfill ● /edit-automation -> adding exponential backoff to step 3 retested. replayed failed runs.
> Stop auto-disabling Shopify on 401s from the sandbox tenant. ● /alert-handling admin_ignore_401_domains += sandbox.shopify.com alert_suppress: 'Stale sandbox token' pattern verified against last 24h history noise dropped from 140 to 3 alerts/day
Traditional platforms wrap APIs in UI. APIANT wraps the entire platform in a Claude-Code-addressable plugin. The difference shows up everywhere.
Integrations that used to take weeks ship in a single Claude Code session. Iteration is conversational.
Skills encode what senior APIANT engineers do. Claude doesn't improvise. It follows vetted playbooks with verification steps.
The plugin separates dev and prod. Claude can inspect prod freely; writes require an explicit deploy skill.
The APIANT Claude Code plugin ships with an APIANT license. Tell us what you are integrating, we set you up, and Claude runs the platform from your terminal.
Your request will be posted to our Discord community, where independent builders who run their own APIANT-powered platforms can see it and reach out to you directly to build the integration.
← BackWe'll take a quick look to learn about your business.
Based on what we found, here are integration products that might be a great fit.
Your integration specialist is ready. Alex already knows your platform and what you're looking to build. Choose how you'd like to connect:
Voice call happens right in your browser. WhatsApp options open the app.
Prefer email? We'll follow up at within 2 hours.
Click the button below to start talking. Alex already has your details.
Your browser will ask for microphone access.
← BackExpect to hear from us within one business day.
Something went wrong. Please try again.