01
// knowledge search_docs(query) → Passage[]
Search your docs
Give an agent read access to your knowledge base so it can cite the answer.
HTTPSecrets
Tool kinds Built on MCPOpen by design
Intelligence, meet execution.
Build, deploy and host MCP servers on one platform, then connect them to any AI client. You bring the logic. We run it.
- Go, TypeScript or PythonOne Lambda per project
- One MCP endpointOAuth gateway
- Free planA month's usage included
Describe a tool, or write one function in Go, TypeScript or Python. We build and test it, run it in your project's own Lambda, and serve it to every AI client through one authenticated endpoint.
How to deploy an MCP serverThree moves.
Three moves.
Zero servers.
Deploy
Describe a tool in YAML, add a run function in Go, TypeScript or Python if it needs code. The platform builds it, tests it and puts it live.
gxmesh pushConnect
Point any MCP client at your project endpoint. One OAuth gateway brings every enabled tool into reach.
one endpointObserve
Follow each invocation from request to result. See outcomes, timing and usage across the project.
invocations
Follow the requestFrom model intent.
From model intent.
To authenticated action.
Every call takes the same six hops. Your keys stay on the server side of the line.
- 01 · MCP request
Request received
tools/call → search-knowledgeAn AI client calls a tool on your project endpoint, carrying its Glxymesh access token, not your secrets.
- 02 · User & client auth
Identity verified
identity → project → permissionsThe gateway checks the user, the client and project access before anything runs.
- 03 · Isolated runtime
Function invoked
invoke → project LambdaYour code runs in your project's own Lambda, with no network but the way out through egress.
- 04 · Keys at egress
Key added
egress → allowed host + keyThe egress gateway checks the host is allowed and adds the key bound to it. Your code never holds it.
- 05 · Data & model auth
Data retrieved
authenticated call → APIThe call reaches your data or model API with its credential, and the answer comes back the same way.
- 06 · Tool response
Result returned
result → gateway → clientA structured result returns to the client, and the call stays visible in the console.
MCP tools and integrationsOne function.
One function.
A new ability for every agent.
Example tools · scroll to explore
02
// data query_db(question) → Row[]
Query a database
Connect to Postgres, MySQL or Redis over TCP and run fixed, parameterized queries, with the connection string kept in project secrets.
SecretsTCP
Connect to a database 03
// workflow create_ticket(summary) → TicketRef
Open tickets from chat
Turn a conversation into a ticket in your tracker, with the right fields filled.
External APISecrets
Keys and network 04
// integration fetch_url(url) → Document
Call any HTTP API
Wrap a third-party API once and every connected client can use it.
External API
Keys and network 05
// authoring propose_tool(request) → Draft
Let an agent build it
Your agent drafts a tool from an API's docs. You review it, approve its keys and publish.
AgentsReview
Build with AI 06
// reporting run_report(period) → Report
Run a report on demand
Package a report as a tool your agent can run and summarise when asked.
FunctionSecrets
Tool kinds MCP infrastructurePowerful by nature.
Powerful by nature.
Controlled by design.
The infrastructure your tools need, with the boundaries your team expects.
A runtime with boundaries.
Each project's code runs in its own Lambda with no direct network. Every outbound call leaves through one gateway that knows which hosts are allowed.
LAMBDA · EGRESS GATEWAYOne endpoint. Every tool.
A single gateway exposes your project's enabled tools to MCP clients, with authentication built in.
STREAMABLE HTTP · OAUTHSecrets stay secret.
Keys are encrypted, scoped to a project and added on the way out. Neither your code nor the client calling your tools ever sees them.
ENCRYPTED · ADDED AT EGRESSYour workspace. Your rules.
Organize tools by project and manage people with role-based access.
OWNER · ADMIN · MEMBER
Connect MCP servers to any AI client
OpenAIClaudeGeminiMistral
Built for your flowFrom your terminal.
From your terminal.
Into the mesh.
- Scaffold locallygxmesh init starts a project; gxmesh new adds a tool you edit in any editor.
- Run it on your machinegxmesh dev serves the project as an MCP server and rebuilds on save.
- Build on the platformgxmesh push builds, tests and deploys it.
$ curl -fsSL https://platform.glxymesh.com/install.sh | bash# Sign in and select your organization
$ gxmesh login$ gxmesh org use# Create a project, then scaffold your first tool
$ gxmesh project create production$ gxmesh project use production$ gxmesh new hello-mesh$ cd hello-mesh# Build and deploy
$ gxmesh pushYour next idea starts here.
PricingStart free.
Full pricing and cost estimator Start free.
Scale when it's real.
Free
$0/ month
No card needed
Try the platform with real tools. Calls stop when the month's usage runs out, so there is never a bill.
- Usage includedA month's worth
- AI credits5 a month
- Hosted tools10
- Time per call10 s
Pro
Most chosen$29/ month
₹2,599 a month in India
For a developer shipping tools to real users.
- Usage includedA month's worth
- AI credits100 a month
- Hosted tools250
- Time per call60 s
Team
$79/ month
₹6,999 a month in India
For a team running tools in production.
- Usage includedA month's worth
- AI credits300 a month
- Hosted tools2,500
- Time per call60 s
A few good questions
Before you build.
What is Glxymesh?
Glxymesh is managed MCP server hosting. Describe tools in YAML or write them in Go, TypeScript or Python; each project becomes an MCP server that we build, host and serve to compatible AI clients through an authenticated gateway.
What is the Model Context Protocol?
The Model Context Protocol (MCP) is an open standard that lets AI applications discover and call outside tools and data in one common way. An MCP server offers tools; an MCP client, such as Claude, ChatGPT, Cursor or VS Code, connects to it and lets the model call them.
How do I deploy an MCP server?
Install the gxmesh CLI, sign in, scaffold a tool with gxmesh new and run gxmesh push. The platform builds it and hosts it for you, and your project's MCP endpoint serves the tools you enable. There is no server for you to run.
Which languages can I use?
Go, TypeScript and Python. A tool that only calls an API needs no code at all: you describe the request in YAML. Each project's code runs in its own AWS Lambda, and every outbound call and database connection goes through the egress gateway.
How do I connect MCP servers to AI clients?
Each project has one MCP endpoint. Any client that supports remote MCP over Streamable HTTP with OAuth, such as Claude Code, Claude Desktop, Cursor or VS Code, can connect to it and sign in through the gateway. The console's Connect page shows the endpoint and each client's setup.
Should I self-host my MCP server or use managed hosting?
Self-hosting means you run the server, TLS, OAuth for remote clients, secrets, scaling and logs. Managed MCP server hosting takes those off your hands: you push tools and get an authenticated endpoint. Self-host when a server must run inside your own network or on a user's machine.
How are API keys and secrets kept safe?
Secrets are encrypted and scoped to a project. The AI client that calls your tools never receives your service or model API keys, and each organization's data is kept separate.
Is there a free plan?
Yes. Free includes a month's usage and 5 AI credits, and calls stop when the usage is used up, so it never bills. Pro is $29 and Team $79 a month (₹2,599 and ₹6,999 in India), each with a month's usage and AI credits included.
Is the Control Surface available?
The Control Surface is a planned operator interface for automation. This platform focuses on deploying and connecting MCP tools; Control Surface integration comes later.