The MCP Gateway that cuts your token bill.
Your agent forgets between sessions, so it re-sends the same context, re-fetches the same data, and re-derives what it already worked out. Every repeat is tokens you pay for. Tulimoa is one endpoint with a memory layer that holds the thread, so your agent repeats itself less.
- EU hosting, Frankfurt
- OAuth, revocable
- Open MCP standard
Agents forget, and forgetting costs tokens.
Tool definitions, re-sent every session.
Federate a dozen MCP servers and every tool schema gets dumped into the context on each run. Your agent pays to re-read the same catalog it already saw.
The same data, fetched twice.
The agent already pulled that customer, that invoice, that order. After a context reset it does not remember, so it calls the tool again and pays for the round trip a second time.
Decisions, re-derived from scratch.
Goals, IDs, and the reasoning behind them fall out of context on a long run. The agent reconstructs what it already knew, one expensive step at a time.
One endpoint. A memory that carries the thread.
Point your agent at one URL. Behind it, Tulimoa federates your MCP servers and runs a memory layer that keeps what matters alive across sessions and context resets.
Point every agent at the gateway and sign in once. Add or remove tools without reconfiguring the agent. It picks them up on the next run.
The gateway captures goals, decisions, and the IDs that flow through it, then hands them back only when they matter, across sessions and after a context reset.
Your connected MCP servers show up behind a single tools list, namespaced and routed. The Tulimoa directory server is wired in from day one.
Connect the gateway once by OAuth over your Tulimoa account. Secrets stay encrypted, used only at the moment of a call, and every connection is revocable.
Less re-sending is the whole point.
Lower cost
Fewer tokens per run means a smaller invoice at the end of the month. The bill is the number we care about.
Faster answers
Less context to read on every step means the model spends less time before it acts.
Fits the window
Carry a distilled thread instead of a raw dump, and long runs stop hitting the context limit.
We are in beta, so we are not putting a number on this yet. The gateway logs what it injects and what agents recall, and we will publish the real savings from beta usage instead of a figure we made up.
Add Tulimoa's MCP server in one line.
The full Gateway is in early beta, but Tulimoa's own MCP server is already live. Add it to your agent and it can search the curated directory right now. Read access needs no account.
claude mcp add --transport http \
tulimoa https://mcp.tulimoa.com/mcp{
"mcpServers": {
"tulimoa": {
"type": "http",
"url": "https://mcp.tulimoa.com/mcp"
}
}
}openclaw mcp add tulimoa \
--url https://mcp.tulimoa.com/mcp \
--transport streamable-httpEndpoint: https://mcp.tulimoa.com/mcp · Streamable HTTP · read access is authless.
Five tools over Streamable HTTP: search the directory, open a listing, list categories, and submit or edit your own with a write-scoped login.
A hand-reviewed catalog of agent-ready software. Free to browse, on the web and over MCP.
Built in Germany, hosted in the EU. GDPR by default, secrets encrypted, connections revocable.
Straight answers.
Is the Gateway live?
It is in early beta. The memory tools and federation are built and the endpoint is up, but it is not self-serve yet: there is no public sign-up flow for it today. You can add Tulimoa's live MCP server now, and join the beta to get Gateway access as it opens up.
How does memory actually cut tokens?
By letting the agent avoid repeats. It recalls an ID or a decision instead of re-fetching or re-deriving it, and a one-time instruction seed survives context compaction so the agent keeps its bearings. Less repeated context means fewer tokens billed.
How much will I save?
We are not going to invent a number. The gateway measures what it injects and what agents recall, and we will publish real figures from beta usage. How much you save depends on how repetitive your agent's work is.
What is MCP?
The Model Context Protocol is an open standard for connecting AI agents to tools. One endpoint, one schema. Tulimoa speaks MCP on both sides: it is an MCP server your agent can call, and a gateway that federates other MCP servers.
Where is it hosted?
In the EU, in Frankfurt. Tulimoa is built in Germany. Secrets are stored encrypted and used only at the moment of a call, and every connection is revocable.
What does it cost?
Browsing the directory and adding the MCP server are free. The Gateway is free during the beta while we learn. Pricing later is usage based, and memory stays free.
Stop paying to re-send what your agent already knew.
Add Tulimoa's MCP server today, and join the Gateway beta to put memory in front of your whole stack.