What does the MCP gateway do?

Agents connect to a Cognocient URL instead of directly to your MCP server. Every tool call is authenticated, checked against a tool policy, scanned for sensitive data, counted against budgets and logged for attribution, then forwarded with credentials only the gateway holds.

How does the tool policy work?

Each server has an optional allow-list and a deny-list. A denied tool call is refused with a JSON-RPC error and never reaches the server, and the tool list the agent sees is filtered so forbidden tools are not even visible. The deny-list always wins.

Are tool arguments stored?

No. Each call is logged with the server, tool name, a hash of the arguments, cost, latency, status and your attribution tags, never the arguments themselves. Arguments are scanned for PII and secrets and can be alerted on, redacted or blocked per your guardrail setting.

Can a tool call cost money?

You can set a per-call price on a server. Paid calls count against budgets and are refused if a blocking budget would be exceeded. Failed calls are not billed.

How do I govern MCP servers with the Cognocient gateway?

Put your MCP servers behind Cognocient: per-server tool allow and deny lists, argument scanning for PII and secrets, per-call pricing, budgets and full attribution for every agent tool call. Business plan.

Agents that call tools are a spend and data-exposure surface of their own. The gateway gives MCP servers the same controls your model traffic already has.

Add a server

Open MCP Gateway → Add server:

Field	Meaning
Name	Becomes the gateway URL: `https://api.cognocient.com/mcp/<name>`
Server URL	Your server's Streamable HTTP endpoint. Public HTTPS only.
Credential header and value	The server's own auth (for example `Authorization: Bearer …`). Encrypted, injected by the gateway. Agents never hold it.
Tool policy	Allow every tool except those denied, or allow only those listed
Cost per tool call	Optional, for paid tools
Scan arguments	Check tool arguments for PII and secrets (on by default)

Use Discover tools to list what the server offers and click tools to build the policy from real names.

Connect an agent

Point your MCP client at the gateway URL and authenticate with your Cognocient key (or a JWT) as the bearer token. The Cognocient key is never forwarded to your server.

# Python MCP SDK
async with streamablehttp_client(
    "https://api.cognocient.com/mcp/github",
    headers={"Authorization": "Bearer <your proxy key>", "X-Cost-Feature": "release-bot"},
) as (read, write, _):
    ...

What is enforced

Authentication, Emergency Freeze, rate limits and budgets: the same as model traffic.
Tool policy: a denied tools/call returns a JSON-RPC error (-32001) and never reaches the server. tools/list responses are filtered, in both JSON and SSE form.
Argument scanning: PII and secrets in arguments follow your guardrail mode: alert, redact before forwarding, or block (-32002).
Paid tools: the per-call price is checked against budgets before the call and recorded only if it succeeds.

Attribution

Each tools/call appears in your call log with provider mcp, the server name, the tool name, a hash of the arguments, cost, latency, status and your X-Cost-* tags, so agent activity sits next to model spend in dashboards and reports. These calls are excluded from retry and context-bloat waste detection, because agents legitimately repeat tool calls.

Denied and blocked attempts are logged too, so you can see what agents tried to do.

Limits

Streamable HTTP transport, one JSON-RPC message per request. Batching was removed from the MCP spec and is refused. The older HTTP+SSE transport is not supported.
Policy applies to tools. Resources and prompts are forwarded unmodified.
Tool results are not scanned; only arguments are.
The server URL must be reachable from the public internet.