MCP Overview

Sprigr Teams supports the Model Context Protocol (MCP), an open standard that lets AI applications connect to external tools and data sources. With MCP enabled, you can manage your entire Sprigr Teams platform - agents, workflows, knowledge bases, marketplace apps, schedules, websites, and more - directly from tools like Claude Code, Cursor, VS Code, and any other MCP-compatible client.

This is not just a chatbot bridge. The MCP server exposes 104 tools that give you full read and write access to the platform, so your external AI client becomes a complete management interface for your Sprigr Teams workspace.

What is MCP?

The Model Context Protocol is an open standard for connecting AI assistants to external systems. Think of it like a universal adapter: instead of building custom integrations for every AI tool, MCP provides a single protocol that any compatible client can use to discover and call tools from any compatible server.

Sprigr Teams acts as an MCP server, exposing your agents, knowledge bases, workflows, apps, and platform management capabilities as tools that external AI clients can use.

How it works

When you connect an MCP client (like Claude Code or Cursor) to your Sprigr Teams workspace:

1. The client discovers every tool your API key’s scope allows (read tools, write tools, or both)
2. You can invoke those tools from within the client’s interface
3. Almost all tools call the Sprigr Teams API directly - no AI tokens consumed, no latency from agent processing
4. Only send_message routes through an agent (because that is the one operation where you want the agent to think and respond)
5. Results are returned to the client and incorporated into the conversation

This architecture means you can list agents, create workflows, search knowledge bases, manage apps, and configure schedules without any agent token cost. The MCP client itself is the intelligence - it reads data through Sprigr Teams, decides what to do, and writes changes back directly.

MCP Client → MCP Worker → Provisioning API → Database
                                (direct - no agent involved, no tokens used)

MCP Client → MCP Worker → Gateway → Agent → AI Response
                                (only for send_message - uses agent tokens)

Supported clients

Sprigr Teams works with any application that supports the MCP standard via Streamable HTTP transport. Popular clients include:

• Claude Code — Anthropic’s CLI tool for developers (recommended - supports remote HTTP servers natively)
• Claude Desktop — Anthropic’s desktop application
• Cursor — The AI-powered code editor
• VS Code — With MCP-compatible extensions
• Any MCP client — Any application that implements the MCP Streamable HTTP client specification

Available tools via MCP

Sprigr Teams exposes 104 tools through MCP: 30 read tools and 74 write tools. Which tools are available depends on the API key’s scope — a read-only key sees the read tools, a write-only key sees the write tools, and a read & write key sees everything.

Read tools (30 tools)

Read tools fetch data from your platform without consuming any agent tokens.

Agents, teams, and people

Tool	Description
list_agents	List all agents with name, slug, status, and role. Use the slugs from this list to call other tools.
get_agent_info	Get detailed agent configuration including persona, model tier, and integrations.
list_teams	List all agent teams in the company.
get_team	Get team details including members and routing rules.
list_users	List all team members.

Workflows and executions

Tool	Description
list_workflows	List all workflows, optionally filtered by project or mode (DAG workflows, code-first Scripts, or both).
get_workflow	Get workflow details including steps and configuration.
list_executions	List recent workflow executions across all workflows, with optional status filtering.
get_execution	Get full detail for a single execution, including per-step status, output, error, and timing.
list_failed_executions	List recent failed executions with the failing step and error for each — the debugging entry point.
list_workflow_versions	List published versions of a workflow with the active version pointer and per-version notes.

Apps, webhooks, and integrations

Tool	Description
list_apps	List all marketplace apps available to the company.
get_app	Get app details including versions and installation status.
list_webhooks	List all configured webhooks.
list_integrations	List all connected integrations (Gmail, Gorgias, GitHub, etc.).

Company overview

Tool	Description
get_usage	Get company usage statistics (messages, tokens, costs).
get_plan	Get current subscription plan and limits.
get_dashboard	Get dashboard overview: active agents, messages today, monthly spend, escalations.
list_schedules	List all scheduled tasks.

Knowledge and conversations

Tool	Description
search_knowledge	Search an agent’s knowledge base using natural language. Each agent has its own knowledge index.
list_knowledge	List knowledge base entries.
list_conversations	List recent conversations for a specific agent. User-bound keys see their own conversations; admin/service keys see every conversation for the agent.
list_threads	List inbox threads (support tickets, customer conversations). User-bound keys see threads they have access to.

Write approvals

Tool	Description
list_mcp_write_approvals	List pending platform-write access requests filed by deep-task specialist agents. Mirrors the portal Approvals page.
decide_mcp_write_approval	Approve or reject a pending write access request (owner/admin only).

Websites (tenant hosting)

Tool	Description
website_status	Get a website’s status, active deployment, and recent builds.
get_build_log	Fetch the captured stdout/stderr from a build, for diagnosing failures.
get_build_source	Recover the source files a build was deployed from.
get_website_logs	Fetch recent platform-side log events for a website (4xx/5xx responses, deploy events).
query_website_db	Run a read-only SELECT against a website’s per-site database.

Write tools (74 tools)

Write tools create, update, and manage platform resources.

Messaging

Tool	Description
send_message	Send a message to an agent and receive its AI response. This is the only tool that consumes agent tokens.
poll_message	Fetch the reply to a send_message that was still processing when it returned. Free and bounded.

Agents, teams, and people

Tool	Description
create_agent	Create a new AI agent with name, slug, persona, and model tier.
update_agent	Update an existing agent’s configuration (persona, status, model tier, etc.).
create_team	Create a new agent team for collaborative task routing, optionally with an orchestrator agent.
add_team_member	Add an agent to a team.
remove_team_member	Remove an agent from a team.
invite_user	Invite a new team member to the company.
notify_chat	Post a message into team chat as an agent — a direct message to a person or a post into a group or channel.
create_team_room	Create a team-chat group or channel as an agent.

Workflows, scripts, and projects

Tool	Description
create_workflow	Create a multi-step workflow with agent assignments, code steps, approval gates, and manual, event, or schedule triggers.
update_workflow	Update an existing workflow’s steps or configuration.
update_workflow_step	Patch one step of a workflow in place without resupplying the others.
delete_workflow	Permanently delete a workflow and its executions.
start_workflow	Start a workflow execution with optional input data.
approve_execution	Approve or reject a workflow step that requires human approval.
publish_workflow_version	Publish the current draft as a new immutable workflow version and make it live.
rollback_workflow_version	Swap the active version back to a previously published one.
create_script	Save reusable, re-runnable code as a Script (a workflow with one code step).
list_scripts	List saved Scripts.
list_projects	List the company’s projects (feature containers grouping related workflows).
get_project	Get one project with its member workflows and how they interact.
create_project	Create a project.
update_project	Update a project’s name, description, colour, or status.
delete_project	Delete a project (member workflows detach rather than being deleted).

Marketplace apps

Tool	Description
create_app	Create a new marketplace app with code that runs in a V8 isolate sandbox.
install_app	Install a marketplace app with specific scopes and agent restrictions.
uninstall_app	Remove a marketplace app from the company.
configure_app_install	Reconfigure an existing install in place: agent scoping, config, and per-install secrets.
upgrade_app	Upgrade the company’s install of an app to the latest approved version.
force_install_upgrade	Force an install upgrade when the automatic publish fan-out missed it.
delete_app	Permanently delete an app the company publishes, across all tenants (publisher only).
publish_version	Publish a new version of an existing marketplace app.
publish_app_listing	Make an app you publish available to every tenant in the marketplace.
share_app	Generate an invite/share link for an app you publish.
unshare_app	Revoke an app’s share link.

Cross-tenant grants

Tool	Description
list_pending_grants	List cross-tenant grants awaiting a decision from your company.
approve_cross_tenant_grant	Approve a pending grant so the capability becomes live.
reject_cross_tenant_grant	Reject a pending grant.
confirm_cross_tenant_grant_revocation	Confirm a publisher’s intent to drop a grant.
keep_cross_tenant_grant_on_revocation	Keep a grant live even though the publisher dropped it.
pause_cross_tenant_grant	Temporarily pause an active grant without revoking it.
resume_cross_tenant_grant	Resume a paused grant.

Webhooks

Tool	Description
create_webhook	Create a webhook endpoint. Incoming webhooks route to an agent, workflow, code, or index; outgoing webhooks fire on events.
update_webhook	Update a webhook’s destination, auth, transform, or status without changing its URL.
delete_webhook	Permanently delete a webhook and its URL.

Knowledge and search indexes

Tool	Description
write_knowledge	Write or update entries in the target agent’s private knowledge index (the same index search_knowledge reads).
create_knowledge	Create a new knowledge base entry (document, FAQ, etc.).
sprigr_create_index	Create a new index with private, company, or public scope.
sprigr_attach_index	Register an existing index in the company metadata table.
sprigr_list_my_indexes	List indexes registered for the company, with 30-day usage stats.
sprigr_delete_index	Delete an index (creator only); associated read keys are revoked.
sprigr_create_public_key	Mint a read or write API key on a public-scoped index.
sprigr_list_keys	List active keys for an index (metadata only, never the secret).
sprigr_rotate_key	Rotate an index key — new secret, same access, old key revoked.
sprigr_revoke_key	Revoke an index key immediately.
sprigr_update_key_label	Rename an index key.

Collections (structured records)

Tool	Description
collection_define	Declare a typed, faceted collection for structured records (asset registers, inventories, contacts, catalogues).
collection_ingest	Upsert a batch of typed records, validated against the schema and deduplicated by key.
collection_ingest_from_table	Ingest parsed spreadsheet/CSV rows via a declarative column-to-field mapping.
collection_query	Query records with full-text search, facet filters, facet counts, sort, and pagination.
collection_reconcile	Read-only diff of a collection against another source: added, removed, changed, unchanged.
collection_describe	Introspect a collection’s schema, object count, and facet values.
collection_history	Read the append-only change history for a collection.

Schedules

Tool	Description
create_schedule	Create a scheduled task — message, reminder, pipe, or script — on a cron or a one-off timestamp.
update_schedule	Update or toggle a scheduled task.
delete_schedule	Remove a scheduled task.

Websites (tenant hosting)

Tool	Description
create_website	Provision a new website with a live starter page on a *.sites.sprigr.com URL.
delete_website	Soft-delete a website (kept for 30 days before purge).
enable_wordpress	Convert a website into a live WordPress + WooCommerce store.
start_build	Enqueue a build for a website from a files map.
rollback_deployment	Flip a website back to a previous deployment.
add_domain	Attach a custom hostname to a website.
set_website_env	Set or rotate an environment variable or secret for a website.

Keys, scopes, and permissions

Every MCP API key has a scope — Read only, Write only, or Read & Write — which controls which of the tool groups above the client can see and call.

Keys created in the portal are also linked to the user who created them. The key carries that user’s identity and role on every request:

• User-bound keys see their own conversations and inbox threads through list_conversations and list_threads — the same view that user has in the portal.
• Admin and owner keys see all conversations, threads, and company data.
• A few sensitive tools, such as decide_mcp_write_approval, are enforced server-side as owner/admin only.

The user’s role is looked up on every request, so a role change (for example, a member promoted to admin) takes effect on the key automatically. MCP access never exceeds what the user can do through the web portal — you cannot gain broader access by using MCP instead of the dashboard.

Use your own AI tokens to build infrastructure

This is one of the most powerful aspects of the MCP integration: when you connect Claude Code, Cursor, or Claude Desktop to Sprigr Teams, the AI reasoning happens on your side using your own Claude tokens (from your Anthropic subscription or IDE licence). Sprigr Teams just serves as the data layer.

Here is why that matters:

• Every tool except send_message is a free API call. When Claude Code calls list_agents, create_workflow, or install_app, it hits the Sprigr provisioning API directly. No Sprigr agent is involved. No Sprigr tokens are consumed. The intelligence is your Claude session - it reads data, reasons about it, and writes changes back.
• Only send_message uses Sprigr agent tokens. This is the one tool where you deliberately want a Sprigr agent to think and respond - for example, asking your Gorgias agent to check for urgent tickets. Everything else is a direct database operation.
• Your AI client does the heavy lifting. When you ask Claude Code to “set up a new agent team with three specialists and a workflow”, Claude plans the work, calls create_agent three times, calls create_team, calls add_team_member three times, and calls create_workflow - all using your Claude tokens for the reasoning and Sprigr’s free API for the execution.

This means you can use the most capable AI models available to you (Claude Opus, Sonnet, or whatever your IDE provides) to orchestrate complex platform setup - without worrying about Sprigr token costs for the planning and decision-making.

Note

Tools like start_workflow and create_schedule are free at call time, but the workflows and scheduled tasks they set up will consume agent tokens later when their agent-driven steps actually run.

Example: Setting up a new department from your IDE

Imagine you are onboarding a new finance department. In Claude Code, you say:

“Set up our finance department. Create three agents - an Accounts Payable agent, an Accounts Receivable agent, and a Financial Reporting agent. Put them in a team called ‘finance-team’. Give the AP agent a persona focused on invoice processing and supplier management. Give the AR agent a persona for customer invoicing and collections. Give the reporting agent a persona for generating P&L summaries and expense reports. Then create a workflow called ‘Monthly Close’ with steps for each agent to compile their data and a final approval gate for the CFO.”

Claude Code reads the instruction, plans the work, and executes it using MCP tools:

1. create_agent × 3 - creates each agent with tailored personas (free)
2. create_team - creates the finance-team (free)
3. add_team_member × 3 - adds each agent to the team (free)
4. create_workflow - creates the Monthly Close workflow with steps and an approval gate (free)

Total Sprigr token cost: zero. All the reasoning was done by your Claude Code session using your own Anthropic tokens. Sprigr just processed the API calls.

Example: Building and deploying a custom app

“Create a marketplace app called ‘invoice_status’ that connects to our Xero API and returns the status of an invoice by number. It should accept an invoice number as input and return the invoice status, amount, due date, and whether it’s overdue. Declare Xero as a required dependency. Use the domain api.xero.com in the allowlist.”

Claude Code writes the handler code, defines the input schema, and calls create_app - all in one shot. Your Claude tokens handle the code generation. Sprigr stores and deploys the app for free.

Example: Building a website dashboard

“Create a website dashboard that shows our team’s daily metrics - active conversations, messages processed today, pending escalations, and monthly spend. Use get_dashboard to pull the data and create a clean HTML page with auto-refresh.”

Claude Code calls get_dashboard to understand the data shape, then calls create_website and start_build to deploy the dashboard - all from your terminal.

Example: Communicating with your agents

When you do need an agent’s intelligence - its persona, integrations, and domain knowledge - use send_message:

“Ask the gorgias-support agent if there are any urgent tickets and have it notify Harry about the status.”

This is the one operation that uses Sprigr agent tokens, because you want the Gorgias agent to actually check Gorgias, apply its customer support expertise, and compose a notification. The cost is one agent conversation turn - the same as if you had typed the message in the portal chat. If the agent is still working when send_message returns, it hands back a message id you can check with poll_message (free) until the reply is ready.

The economics in practice

A typical MCP session might involve 20-30 tool calls to set up infrastructure, search knowledge, and configure workflows - all free. Then 1-2 send_message calls to have agents do actual work. You get the full power of your AI subscription to orchestrate the platform, and only pay Sprigr tokens when you need an agent to think.

Use cases

Full platform management

Manage your entire Sprigr Teams workspace from your IDE. Create agents, configure workflows, install apps, and invite users - all without opening the portal or spending agent tokens.

Build apps from your IDE

Write custom marketplace apps in Claude Code or Cursor. Your AI writes the handler code, defines the schema, and deploys it to Sprigr - all using your own Claude tokens for the code generation.

Automated workflows

Build and trigger complex workflows from your terminal. Create a workflow, assign agents to steps, start an execution, and approve results - all through MCP tools with zero Sprigr token cost.

Knowledge management

Write to and search knowledge bases from your external AI client. Push documentation, ingest structured records into collections, and keep your agents’ knowledge up to date without opening the portal.

Websites and hosting

Provision websites, deploy builds, attach custom domains, and even spin up WordPress stores - straight from your MCP client.

Agent conversations

Send messages to any agent and get AI-powered responses. Check urgent support tickets, ask your accounting agent for invoice status, or have your scheduling agent find available slots - the only operation that uses Sprigr tokens.

Example: checking urgent tickets from your IDE

From Claude Code or any MCP client, you can ask:

“Send a message to the gorgias-support agent asking if there are any urgent tickets, and have it notify Harry.”

This uses one send_message call (the only tool that costs Sprigr agent tokens). The Gorgias Support agent checks for urgent tickets, finds what needs attention, and notifies Harry - all triggered from your terminal.

Getting started

To connect an MCP client to your Sprigr Teams workspace, you will need:

1. An API key — Generated from the MCP page in the Sprigr Teams portal. See API Keys for instructions.
2. The server URL — https://mcp.team.sprigr.com/mcp
3. Client configuration — Depends on your client. See the API Keys page for setup instructions for Claude Code, Claude Desktop, and other clients.

Tip

For Claude Code (CLI), add the server with a single command:

claude mcp add --transport http "my-sprigr" https://mcp.team.sprigr.com/mcp --header "Authorization: Bearer sk_mcp_YOUR_KEY_HERE"

Next steps

• API Keys — Create and manage API keys for MCP access.
• Knowledge Bases — Set up knowledge bases that MCP clients can search.
• Creating Agents — Create agents that MCP clients can interact with.
• Marketplace Apps — Build and install custom tools that agents can use.