Every agent. Every tool. One gateway.

Replace scattered MCP servers with one managed endpoint for the whole company. Set the rules centrally, inspect every call, and revoke access when needed.

Gmail, GitHub, Slack, Notion, HubSpot, Linear, Google Drive, Google Sheets, Salesforce, Jira, Figma, Airtable, Zendesk, Shopify and hundreds more connect through the gateway.

1,000+ apps available
Trusted by
Google

The problem

Agent adoption moved faster than governance

mcp-server-prodeng
slack-tools (fork)no owner
crm-bridge-v0.3no owner
local-mcp — Dan's laptopno owner

01

Server sprawl

Every team spins up its own MCP servers. Nobody knows what's running, or with whose credentials.

.env.local SLACK_TOKEN=xoxb-9f…
config.json "apiKey": "sk-live-4e…"
notes.txt crm pass → shared drive
3 copies · owner left in March

02

Scattered secrets

Tokens live in config files across laptops. Offboarding means hoping you found them all.

14:02CRM_UPDATE_DEALserver-a
14:05SLACK_POST_MESSAGEserver-b
14:11— record deleted —which agent?
14:15SHEETS_APPEND_ROWserver-a

03

Zero visibility

When an agent does something in prod, there's no unified record of who, what, or why.

How Spinrun solves it

Set the rules once. Every agent follows them.

Servers, access, credentials, and activity — managed in one place and enforced on every call.

01

Only the tools you connected

Agents see the toolkits your workspace authorized and the actions your rules allow — not every server someone once spun up.

Connected means accountable

One shared endpoint

An owner on every toolkit

Connect from the dashboard

Disconnect in a click

Gateway8 toolkits · 3 teams
https://spinrun.ai/mcp
Gmail· OfficeConnected
GitHub· EngineeringConnected
Salesforce· SalesConnected
Slack· OfficeConnected
Notion· OfficeConnected

02

Rules per app, by risk

Allow reads, hold back writes, block anything destructive — set per connected app and checked on every call, not suggested in a prompt.

Checked on every call

Read / write / destructive tiers

Set per connected app

Applies to every agent

Blocked calls are logged

Access policieschecked on every call
ToolkitEngOfficeSales
Gmailread + writeread onlyblocked
GitHubread onlyread + writeblocked
Salesforceblockedread onlyread + write
Slackread + writeread + writeread only

03

Centralized auth

OAuth and API tokens managed in one place, refreshed before they expire and revocable in a click. Agent configs stay clean.

Agents never hold a token

Managed OAuth flows

Keys in an encrypted vault

Refreshed before expiry

Revoked in one click

Credentialsencrypted vault
Salesforce · OAuthActive
6 of 36 scopes granted · refresh: automatic

Lifecycle

14:32Token refreshed automatically
11:04Agent used CRM_GET_LEAD
09:12alex@… offboarded, access revoked

04

Unified activity log

Tool, action, outcome, duration — every call through the gateway, on a retention window your plan sets.

One view for every investigation

User, tool, action, outcome

Blocked calls included

Filter by team or app

Retention set by plan

Activityall servers · all teams
GITHUB_CREATE_PRmira@co · 14:22
CRM_GET_LEADalex@co · 14:28
GMAIL_SEND_EMAILdana@co · 14:41
SLACK_DELETE_CHANNELalex@co · 15:03Blocked
Destructive action · rule: Slack — writes only · payload not stored

How it works

Same agents, one control point

Without a gateway

  • Agents wired straight to separate servers
  • Auth configured connection by connection
  • Tool catalogs eat agent context
  • No shared record of what ran

With Spinrun MCP Gateway

  • One governed endpoint for every agent
  • Identity and policy checked centrally
  • Only relevant tools enter context
  • Every call lands in one audit log

Built for production

Scales past the demo

Big catalogs, big payloads, many clients — the gateway absorbs the load so agents stay predictable.

01

Tools loaded by intent

Agents search in plain language and load only the tools that match. Context stays small no matter how large the catalog grows.

Resolved at call time

Plain-language search

Only matches load

Small context at any size

Steady at 1,000+ toolkits

Tool search4 of 1,000+ loaded
Agent“Triage today's error spikes and open tracker issues”

Loaded into context

SENTRY_LIST_ISSUES
List unresolved issues in a project
SENTRY_GET_EVENT
Get error detail and stack trace
LINEAR_CREATE_ISSUE
Create an issue in a team
SLACK_POST_MESSAGE
Post to a reachable channel

02

Heavy work runs in a sandbox

Big responses execute in an isolated sandbox with a filesystem. The agent reads back a compact summary, not a raw dump.

Big payloads stay out of context

Isolated runtime

Files persist per session

Compact summaries back

Raw dumps stay behind

Sandboxpython · isolated
issues = run("SENTRY_LIST_ISSUES")
ranked = rank(issues)          # 1,284 rows
for e in ranked["P0"]:
    run("LINEAR_CREATE_ISSUE", e)
run("SLACK_POST_MESSAGE", summary)

Returned to agent

47 triaged · 5 P0 · 5 issues opened 1.2 KB

left on the sandbox filesystem: sentry_issues.json · 202 KB

03

Scoped keys per team

One endpoint, many identities. Each workspace key is scoped to a team and revocable on its own, so cutting one off breaks nothing else.

A blast radius of one team

A key per team or pipeline

Scoped to allowed apps

Limits per key

Revocable on its own

Workspace keysone endpoint, scoped keys
https://spinrun.ai/mcp
sk-eng-…f41· Engineeringreads + writes
sk-sales-…a08· SalesCRM only
sk-ci-…9c2· CI pipelineread only
Revoking a key cuts one team off — nobody else notices.

04

Any client, any model

Claude, Cursor, n8n, Zapier — anything that speaks MCP connects the same way. Swap models without rewiring a single tool.

Nothing to rewire

Claude, Cursor, Codex

n8n and Zapier

Any MCP client

Models swap freely

Connected clientsanything that speaks MCP
Claude
n8n
Zapier
+Cursor, Codex, yours…
Same tools on every model:

How it compares

More than a proxy. More than a catalog.

Forwarding calls is the easy part. The job is owning what happens on them — the apps, the access, and the record, under one set of rules.

Plain proxyToolkit catalogSpinrun MCP Gateway
Managed app catalog
you wire each app
1,000+ toolkits
One endpoint for every agent
Action-level access rules
server-level at best
down to one blocked call
OAuth handled for you
refreshed + revocable
One audit trail
per-plan retention
Context-sized tool loading
search, then load
Who maintains it
You
Partly you
We do

Enterprise ready

Built for your security review

Tenant isolation

Row-level tenant isolation in the database

An encrypted vault

Credentials held in an encrypted vault, never in config files

Retention you control

Per-plan log retention, enforced by a nightly purge

Enforced, not suggested

Read / write / destructive rules enforced at the gateway

SOC 2 & ISO 27001-grade practices

Audit-friendly logging, retention controls, and least-privilege defaults across the platform.

Two ways to sign in

OAuth for clients that support it, workspace API keys for the rest — both scoped, both revocable.

Use cases

One gateway, every department

Office Work

Connect Claude or ChatGPT to the apps you already use — Gmail, Slack, Notion, Calendar, and 1,000+ more. One prompt, real work, done.

Explore

Sales & Revenue

Enrich a lead, draft the follow-up, book the meeting, brief yourself for the call — across Salesforce, HubSpot, Gmail, LinkedIn, and 1,000+ more tools. One prompt, deal momentum.

Explore

Marketing & Growth

Launch campaigns, publish social, monitor SEO, pull funnel metrics, and nurture leads — across HubSpot, Mailchimp, Google Analytics, LinkedIn, and 1,000+ more tools. One prompt, growth compounding.

Explore

Product & Design

Synthesize research, iterate designs, write specs, plan sprints, and track metrics — across Figma, Linear, Notion, PostHog, and 1,000+ more tools. One prompt, less context-switching, more shipping.

Explore

Customer Support

Triage the queue, draft the reply, escalate to engineering, and keep the CSM synced — across Zendesk, Intercom, Salesforce, Slack, and 1,000+ more tools. One prompt, every ticket moving.

Explore

Engineering & DevOps

Triage, review, deploy, respond, and retro — across GitHub, Linear, Sentry, Datadog, PagerDuty, and 1,000+ more tools. One prompt, less busywork, more shipping.

Explore

HR & Recruiting

Screen candidates, schedule interviews, send offers, and run onboarding — across Ashby, Lever, BambooHR, Deel, and 1,000+ more tools. One prompt, less admin, more hires.

Explore

Finance & Ops

Categorize expenses, reconcile payouts, approve POs, and run the close — across Stripe, NetSuite, Brex, Xero, and 1,000+ more tools. One prompt, numbers that tie out.

Explore

E-commerce

Manage inventory, triage support, run nurture flows, and tune your ads — across Shopify, Klaviyo, Gorgias, Google Ads, and 1,000+ more tools. One prompt, every order moving.

Explore

Content & Media

Research, generate, schedule, publish, and measure — across YouTube, LinkedIn, Typefully, ElevenLabs, Canva, and 1,000+ more tools. One prompt, the content calendar runs itself.

Explore

IT & Security

Triage alerts, provision access, investigate incidents, and prep audits — across PagerDuty, Datadog, Auth0, Cloudflare, and 1,000+ more tools. One prompt, systems stay up, evidence stays ready.

Explore

FAQ

Frequently asked questions

Put one gateway in front of every tool

Give agents a secure foundation across every tool your company already runs on.

Before anything loads

We would like to switch on analytics, so we can see which pages are worth keeping. Nothing has loaded yet and nothing will until you choose. The cookies that keep you signed in and remember your language are not part of this.

Read the Cookie Policy