Platform · AI governance & security

One endpoint between your agents and every tool they call.

Your agents talk to Meandr instead of to each tool server directly. Meandr checks every call against your rules, keeps rate limits honest, pauses anything risky for a human, and writes it all down. You don't touch your agent's code — and there's no proxy or sidecar for you to run.

Get started See how it works

Everything in one place.

Servers, policies, live traffic, approvals, and cost — the whole control plane between your agents and their tools, on one screen.

Meandr dashboard — screenshot placeholder
Servers, policies, traffic, approvals and cost — one place.

Right now, nothing sits between your agents and your tools.

Every agent holds its own keys and dials each tool server directly. That's fine until it isn't.

Nothing in the middle

With every agent connected straight to every server, there's no single place to see what they can reach — or to change your mind later.

One bad call is all it takes

A confused agent can drain an API key, run up a bill, or do something you can't undo. There's no "ask a human first" for the calls that actually matter.

You can't prove what happened

When someone asks what an agent did last week, provider logs won't tell you which agent, under which rule, with what arguments. That's a problem the day you need an answer.

How it works

One endpoint. Every call, checked.

Your agent connects to Meandr the same way it would connect to any MCP server. Everything below happens on the way through — the agent never notices.

01

Connect

One URL, one catalog.

Point your agents at a single Meandr URL. Add your tool servers once — Meandr merges them into one catalog your agents see as a single server. No SDK, no code changes.

You set up: your servers, agents, and tokens — in the dashboard.

02

Check

Before anything runs.

Every call is checked against your rules first — allow it, block it, or hold it for a human. Rules can look at the actual arguments, not just the tool name. Rate limits apply across your whole fleet at once.

You set up: rules (allow / block / ask a human) and rate limits.

03

Run & log

Then written down.

Approved calls go through to the right server, signed with the right credentials. Blocked calls never leave Meandr. Either way it lands in one audit log — who called what, when, and which rule decided.

You set up: upstream servers, credentials, and what to keep.

One place to say what your agents may do.

The same discipline you already put on your HTTP traffic — a gateway and a WAF — but for the calls your agents make.

No changes to your agents

Your agents keep speaking plain MCP. Swap the address they call for one Meandr URL — that's the whole integration. Nothing new to install on their side.

Rules that read the real call

Match on the actual arguments, not just the tool name. "Auto-approve refunds under $100, ask a human above it" is a rule you can actually write — the only sane way to handle catch-all tools.

A human in the loop when it counts

Hold risky calls for a quick yes/no — and for clients that support it, right inside the agent's own chat, confirmed with a one-time code. Approve and it runs; deny and it never does.

Connect anything — even AWS

Bring your servers however they sign in — Meandr handles the auth for you. API keys, custom headers, basic auth, OAuth, mTLS or AWS SigV4 — it signs each upstream request, so an AWS-hosted MCP server works the moment you add it. No sidecar, no proxy to run.

Rate limits that actually hold

Caps enforced across your whole fleet — not per-box counters that quietly add up to something much bigger. Set them per agent, per project, and per server at the same time.

Every call on the record

Allowed, blocked, or held — each call is one line you can search: who, what, when, which rule. Want the full request and response too? Turn on encrypted capture per rule. Off by default.

Audit log / live traffic — screenshot placeholder
Every call checked against your rules — who called what, and which rule decided.

Questions you'd ask anyway.

Do I have to change my agents?
No. Your agents keep speaking plain MCP — you just point them at Meandr's URL instead of at each tool server. Nothing to install, no SDK, no code change.
How does Meandr sign in to my tool servers?
You hand Meandr the upstream credentials once and it authenticates every call for you, whatever the server expects — API keys, custom headers, basic auth, OAuth, mTLS or AWS SigV4. They're stored encrypted, and your agents only ever hold a Meandr token you can revoke in one click.
Can I connect AWS-hosted MCP servers?
Yes, out of the box. AWS servers expect AWS SigV4-signed requests; Meandr signs them for you. Point it at the server and it works — there's no proxy or sidecar for you to stand up and babysit.
Does Meandr host my tool servers?
No. It sits in front of the servers you already run or subscribe to — your own, or a vendor's hosted MCP. Meandr governs the calls; it doesn't replace the tools.
Is this a workflow builder or a model router?
Neither. Meandr doesn't chain steps together and has nothing to do with which model your agent uses. It governs individual tool calls — that's the whole job.
What happens to a blocked call?
It stops at Meandr and never reaches the tool server — nothing runs. The decision and the rule behind it show up in your audit log.

Put a control plane between your agents and your tools.

Point your agent at one endpoint and you're governed in minutes — no code changes, no proxy to run.

Get started

Questions? Email us