Zero-token MCP execution for autonomous agents

Control. Automate. Deliver.

Control every action, automate the routine, deliver what matters. A local gateway that governs what your agents can do — and a pipeline engine that collapses dozens of tool calls into one clean, validated answer.

Expose only the right tools
Deny-by-default visibility
Human-in-the-loop approval
For sensitive, high-risk actions
Argument policies
Allow-lists and value restrictions
Audit without secrets
Every call recorded, no raw argument values
Drift protection
Schema changes block tools until policy is updated
agent autonomous call MCPToolGate blocked you approved args policy id=1004 ✓ id=5555 ✕ pass chat db cloud your MCP servers TPE Tool Pipeline Engine acquire · map · validate merge · compress one result ↑ $22.02 model · 262 turns $0.93 engine · 12 turns 24× cheaper
01

One execution platform. Two layers. One dependency.

Everything runs through MCPToolGate. The gate governs what agents can touch; the pipeline engine consumes that surface to collect and compress data. Run the gate alone as a security layer — the engine only ever works on top of it.

AI Agent (direct tool calls)
MCPToolGate
Security & execution layer
Tool visibility
Human approval
Argument policies
Audit
no secrets
Drift protection
Tool Pipeline Engine
Automation layer
  • Multi-tool workflows
  • Data collection
  • Transformation
  • Validation
  • Context compression
MCP servers (your tools)
gitchatissuesclouddatabasecrm
02

Four fates for every tool invocation.

All requests pass through the same rules. What happens to a call depends only on the policy you set — the agent just sees the outcome.

1Tool not visible
If a tool isn't exposed, the agent can't call it — it never sees it.
agent delete_database() not visible
2Approval required
Sensitive actions are held for human approval via your channels — Telegram, Slack, anywhere.
agent transfer_money() approval execute
3Arguments rejected
Arguments are validated against allow-lists and policies before the call runs.
agent search(id=5555) policy rejected
4Successful execution
Policy passed, tool executed, logged for audit — clean and traceable.
agent search(id=1004) policy execute audit
03

Tool Pipeline Engine: how it works.

Declarative schema, not code — an agent can author it from a skill and register it with one tool.

1
Acquisition
Call the right tools with parameters. Delta logic uses a watermark to fetch only new data, and respects strict scope boundaries.
2
Field mapping
Lift, rename and transform fields. Merge sources, convert formats. Nothing is lost silently — unmapped_policy: report.
3
Validation
Validate the output against a contract. Required fields and types are guaranteed, or the result is flagged.
Declarative workflow
git · chat · issues · mail · database · …
Extract & merge
collect raw responses
Transform & map fields
rename · convert · extract
Validate & check quality
contract & data quality
Single normalized result
ready for the agent
04

Zero-token execution.

The engine runs the whole tool chain — fetch, merge, transform — without a single model token. The agent only pays for the final answer and for the sources that genuinely need reasoning. Measured on a live assistant wired to 19 MCP servers exposing 200+ tools — code, issues, chat, mail, docs and calendar — same workload before and after.

Model walks the tools
19 sources · 200+ tools
the model reads every one
262 reasoning turns
5.77M
tokens processed · 208k output
$22.02
per full run
Engine walks the tools
19 sources · 200+ tools
collected mechanically, zero tokens
12 reasoning turns
0.40M
tokens processed · 11k output
$0.93
per full run
24×cheaper
Same workload, measured before and after — the model stopped doing mechanical work:
Reasoning turns
26212
22× fewer
Output tokens
208k11k
20× fewer
Tokens processed
5.77M0.40M
14× fewer
Cost per run
$22.02$0.93
24× cheaper
Faster execution Lower costs Better decisions Reliable results
05

The big picture.

Secure execution. Automated workflows. Smarter agents. Whether the agent calls a tool directly or runs a whole pipeline, it goes through the same governance layer.

AI Agent
↓ direct tool callspipeline workflows ↓
MCPToolGate
governance & execution layer
Tool visibility
Human approval
Argument policies
Audit
no secrets
Drift protection
MCP servers (your tools) — git · chat · issues · cloud · crm · database · …
You stay in control
Only the right tools, under the right rules.
Routine is automated
Collect, transform and compress without expensive model calls.
Agents get what matters
Clean, validated data. Better answers.

Stop paying your best model to be a courier.

One local binary. Pure Go, no runtime, cross-platform. Point it at your MCP servers, choose what's exposed, write a schema, and hand your agent signal instead of sprawl. MCPToolGate is in active development — join the waitlist to get early access first.

No spam. One email when early access opens.