N°00 — THE CONNECTION LAYER

One key to every
tool your AI touches.

Lumen is the universal connection layer for AI agents. One endpoint hands Claude, Cursor, Codex — any agent — every tool, model and API it will ever need. No keys to manage. No SDKs. Pay only for what it uses.

Paste into Claude → Connectors. Live in under a minute.

api.lumen.dev/mcplive
Wallet · this month
$2,847.18
cap $5,000
typecalllatencycost
INFERclaude.opus-4.6103ms$0.0002
CALLgmail.send67ms$0.0027
CALLlinear.create_issue173ms$0.0009
QUERYpostgres.select37ms$0.0015
CALLslack.post_message142ms$0.0008
CALLstripe.create_refund21ms$0.0027
4,200+
Tools & APIs
1
Connection
38ms
Median overhead
$0.00
To start

Connects every agent to everything they run on

Claude logoClaude
Claude Code logoClaude Code
Codex logoCodex
Cursor logoCursor
Gemini CLI logoGemini CLI
DDevin
Windsurf logoWindsurf
CCline
Claude logoClaude
Claude Code logoClaude Code
Codex logoCodex
Cursor logoCursor
Gemini CLI logoGemini CLI
DDevin
Windsurf logoWindsurf
CCline
GitHub logoGitHub
Linear logoLinear
Vercel logoVercel
Sentry logoSentry
Postgres logoPostgres
Snowflake logoSnowflake
Stripe logoStripe
Slack logoSlack
Gmail logoGmail
Notion logoNotion
Figma logoFigma
Salesforce logoSalesforce
HubSpot logoHubSpot
Jira logoJira
Drive logoDrive
Calendar logoCalendar
PagerDuty logoPagerDuty
Datadog logoDatadog
Airtable logoAirtable
Shopify logoShopify
Twilio logoTwilio
Cloudflare logoCloudflare
GitHub logoGitHub
Linear logoLinear
Vercel logoVercel
Sentry logoSentry
Postgres logoPostgres
Snowflake logoSnowflake
Stripe logoStripe
Slack logoSlack
Gmail logoGmail
Notion logoNotion
Figma logoFigma
Salesforce logoSalesforce
HubSpot logoHubSpot
Jira logoJira
Drive logoDrive
Calendar logoCalendar
PagerDuty logoPagerDuty
Datadog logoDatadog
Airtable logoAirtable
Shopify logoShopify
Twilio logoTwilio
Cloudflare logoCloudflare
01The integration tax

Every new tool used to cost you a week.

Agents are only as capable as the things they can reach. Wiring up each one by hand is where the magic quietly dies. Lumen deletes that work.

Wiring it up yourself

  • A new MCP server to host for every single tool
  • OAuth flows, API keys and tokens scattered everywhere
  • Secrets baked into prompts the model can read
  • Nineteen invoices, zero idea what your agent actually spent
  • No spend caps — one runaway loop drains the account
  • Rebuild and redeploy every time you add a capability

With Lumen

  • One MCP endpoint. Paste it once, into any agent
  • Credentials held, scoped and rotated in an isolated vault
  • The model never sees a secret — ever
  • One pre-funded wallet, metered to the cent, live
  • Hard caps per agent, per tool, per day
  • Add a provider and it's callable instantly — no redeploy
02Live in three steps

From zero to omni-capable in under a minute.

0101

Connect once

Paste a single MCP endpoint into Claude, Cursor, Codex or any agent. No SDKs, no per-tool OAuth dance, no credential vault to babysit.

lumen connect --agent claude
0202

Fund your wallet

Top up once. Lumen meters every call to the cent and pays each provider on your behalf. One invoice instead of nineteen.

lumen wallet fund $50
0303

Call anything

Your agent discovers and calls every tool, model and API the moment it's online. Permissions, rate limits and spend caps enforced at the edge.

agent → lumen → 4,200+ tools
03The control plane

Watch your agent reach across your whole stack.

Every call routed, authenticated, timed and metered through one endpoint — with a signed ledger you can actually trust.

agent · lumen session
connected
userTriage today's Sentry errors and file Linear tickets for the P1s.
Wallet•••• 4291
Balance
$2,847.20
Calls metered
48,213
Per request
~$0.0004
Live spend by provider
Claude64%
GitHub18%
Stripe9%
Linear5%
Other4%
04One catalog

Every tool your agent needs. Already connected.

Browse a slice of the 4,200+ providers reachable through Lumen. Search, filter, and your agent can call any of them the moment it's online.

GitHub logo
GitHub
Dev
Linear logo
Linear
Dev
Vercel logo
Vercel
Dev
Sentry logo
Sentry
Dev
Postgres logo
Postgres
Data
Snowflake logo
Snowflake
Data
Stripe logo
Stripe
Payments
Slack logo
Slack
Comms
Gmail logo
Gmail
Comms
Notion logo
Notion
Docs
Figma logo
Figma
Design
Salesforce logo
Salesforce
CRM
HubSpot logo
HubSpot
CRM
Jira logo
Jira
Dev
Drive logo
Drive
Docs
Calendar logo
Calendar
Comms
PagerDuty logo
PagerDuty
Ops
Datadog logo
Datadog
Ops
Airtable logo
Airtable
Data
Shopify logo
Shopify
Commerce
Twilio logo
Twilio
Comms
Cloudflare logo
Cloudflare
Data
+ 4,178 more — and counting.
05Not a proxy — a control plane

Everything between your agent and the world.

One endpoint, every capability

Tools, LLMs, vector stores and raw HTTP APIs all sit behind a single MCP connection. Add a provider and it's instantly discoverable and callable — no redeploy, no new config. The whole stack, one surface.

routingauthrate-limitmeteringledgercaching

Zero key management

Credentials are held, scoped and rotated in an isolated vault. Your agent never sees a secret.

Per-request metering

Pay for the exact calls your agent makes. Sub-cent precision, live in the dashboard.

Spend caps & guardrails

Hard ceilings per agent, per tool, per day. Approvals on the calls that matter.

Full call ledger

Every request, argument and response — signed, searchable, replayable. Audit-grade by default.

06One line, any agent

If it speaks MCP or HTTP, it speaks Lumen.

Drop one endpoint into the client you already use. No SDK to learn, no wrapper to maintain. The same connection powers your editor, your CLI and your production agents alike.

  • Works in Claude, Cursor, Codex, Gemini CLI & custom agents
  • Bring your own keys or use Lumen-managed credentials
  • Switch models without touching a line of code
# Settings → Connectors → Add custom connector
https://api.lumen.dev/mcp
 
# Your agent now sees every connected tool.
07Pricing

Fund once. Pay for what your agent uses.

No per-seat lock-in, no minimums, no surprise invoices. The meter only moves when your agent actually calls something.

Starter

$0to begin

For tinkering and personal agents. Fund a wallet, pay only for calls.

Start free
  • Pay-as-you-go, cost + 4% per call
  • All 4,200+ tools & models
  • 1 connected agent
  • 7-day call ledger
  • Community support
Most popular

Team

$49/ mo + usage

For teams shipping agents to production with real guardrails.

Start 14-day trial
  • Everything in Starter, cost + 2%
  • Unlimited connected agents
  • Spend caps & approval workflows
  • Bring-your-own-keys
  • 90-day signed ledger + export
  • Priority support

Enterprise

Customtalk to us

For platforms running agents at scale with compliance needs.

Contact sales
  • Volume pricing, flat platform fee
  • SSO, SCIM & role-based access
  • Private vault & VPC routing
  • Unlimited ledger retention
  • SOC 2 report & DPA
  • Dedicated solutions engineer
08Questions

The honest answers.

A self-hosted MCP server is one tool, one process, one set of keys you maintain forever. Lumen is the whole catalog behind one managed endpoint — provisioned, authenticated, rate-limited and metered. You connect once and never touch infrastructure again.

Anything that speaks MCP or can call an HTTP endpoint: Claude, Claude Code, Codex, Cursor, Gemini CLI, Windsurf, Devin, Cline and your own custom agents. One config line each.

You pre-fund a wallet. Lumen meters every downstream call at cost plus a thin platform fee, pays each provider for you, and shows the running total live. No subscriptions, no minimums, no surprise invoices.

That's the point of the control plane. Set spend caps, scope tools per agent, require human approval on sensitive actions, and review a signed ledger of every call. Credentials live in an isolated vault the model never reads.

Yes. Use Lumen-managed credentials for the long tail, or bring your own for the providers you already pay for. Both route through the same endpoint and the same ledger.

Free to start · no card required

Give your agent superpowers in one line.

Connect once. Fund a wallet. Watch your agent reach across every tool, model and API you run on — metered, guarded and yours.