v3.0·MCP-native gateway is live

One sentence.
Your agent becomes 

AgentForge is the living OS for AI agents. Connect Claude, Cursor, Codex or Grok and, in plain English, install state-of-the-art skills, playbooks, souls and guardrails for your industry — or generate brand-new ones, made just for you.

agent.mcp
live

Compatible with every major agent runtime

CLAUDECURSORCODEXOPENCLAWHERMESGROKLANGCHAINREPLITCLAUDECURSORCODEXOPENCLAWHERMESGROKLANGCHAINREPLIT
How it works

Powerful underneath. Stupidly simple to use.

No prompt engineering. No fine-tuning pipelines. Just talk to your agent like a teammate and AgentForge handles the rest — from discovery to installation to continuous evolution.

01

Connect in 30 seconds

One MCP handshake links your agent to AgentForge. No SDKs, no retraining, no DevOps. Works with Claude, Cursor, Codex, Grok and any MCP-compatible runtime.

02

Tell it what you do

In plain English: "I run a cardiology clinic" or "I close enterprise SaaS deals." AgentForge maps your domain and picks the state-of-the-art skills, playbooks, souls and guardrails for it.

03

Install — or generate — on command

Pre-built packages install instantly. Need something unique? Say "create a soul that negotiates like our top rep" and SkillForge generates a custom skill, playbook or soul tailored to your data.

04

Evolves while you sleep

Every interaction feeds the loop. Better versions ship automatically, guardrails tighten, and your agent's Health Score climbs — without anyone touching a config file.

Plain-English commands

If you can write a sentence, you can ship a specialist.

AgentForge does two things at once: it enriches your agent with the best existing packages for your industry, and it creates new ones — skills, playbooks and souls custom-built from your data — on demand.

Industry★ to pin favorites

Clinical-grade specialists with citation-required guardrails.

›_

Make my agent a cardiologist trained on the latest 2026 ESC guidelines.

Installed cardiology-diagnostics@2.1 + medical-guardrails. Health +6.4.

›_

Hematology specialist for a São Paulo clinic, conservative tone, always cite sources.

Installed hematology-specialist@1.3 + cite-required guardrail. Safety +5.6.

›_

Build a triage flow for chest-pain intake at our ER.

Generated chest-pain-triage@0.1 from your last 30 protocols. Latency −30ms.

›_

Block any answer that gives a dose without patient weight.

Generated dose-safety-shield@0.1. 14 unsafe patterns blocked, 0 false positives.

Same goal, different industry

Pick two industries. Watch the same objective become two different agents.

AgentForge doesn't translate — it re-stacks. Same goal, different packages, different soul, different guardrails.

Objective

First-touch routing with the right level of caution.

Industry A
Industry B
Industry AHealthcare
Adapted command

Triage chest-pain intake at our ER, conservative tone, always cite the 2026 ESC guideline used.

What the agent does

Routes to STEMI vs non-STEMI vs musculoskeletal, blocks dosing without weight, escalates ambiguous cases to the on-call cardiologist.

Stack
  • skillcardiology-diagnostics@2.1
  • playbookchest-pain-triage@0.1
  • guardrailmedical-guardrails
Latency −30ms · Safety +6 · 0 unsafe outputsForge live
Industry BSales & SaaS
Adapted command

Triage inbound from our website form: route ICP fits to AE, others to nurture, flag procurement-shaped asks.

What the agent does

Scores ICP fit, detects procurement language, books the right person — never invents pricing.

Stack
  • skillicp-router@1.3
  • guardrailno-pricing-fabrication
+22% reply rate · 0 invented price quotesForge live
The Stack

Four primitives. Infinite agents.

SKILLS

Capabilities, installable.

Cardiology Diagnostics. Growth Hacking. Legal Due Diligence. Discrete units of expertise an agent can install on demand.

cardiology-diagnosticsgrowth-hacking-prolegal-due-diligence
PLAYBOOKS

Workflows that ship results.

Complete operational logic for real-world processes — from enterprise sales motions to medical consultations.

enterprise-sales-flowmedical-consultation-v2startup-validation
SOULS

Personality as code.

Tone, style, principles, decision-making. Give your agent a Steve Jobs soul, a McKinsey soul, or a humanized doctor soul.

steve-jobs-soulmckinsey-consultanthumanized-doctor
GUARDRAILS

Safety, by default.

Block hallucinations, regulatory violations, out-of-domain behavior and unsafe actions before they reach your users.

medical-guardrailsfinance-complianceno-hallucination
SkillForge AI

The genius behind the geniuses.

A meta-intelligence layer installed on every connected agent. It analyzes weaknesses, recommends upgrades, evolves playbooks, validates performance, and ships improved versions to the entire network — automatically.

  • Recommendation engine across 4,000+ packages
  • Automatic version generation from collective feedback
  • Continuous self-assessment & health scoring
  • Anonymous A2A learning across the network
agent-health-score
+8.4 this week
98.3
/100
Precision96
Soul alignment99
Guardrail integrity100
Playbook coverage92
SkillForge recommends
Upgrade enterprise-sales-flow → v1.5.0
+12% close rate observed across 1,284 agents
Continuous evaluation loop

Your stack evaluates itself, every minute, forever.

Five phases run on every primitive in your stack — skills, playbooks, souls and guardrails — without retraining and without downtime. You see the signal, AgentForge acts on it.

What the loop does to each primitive

The same five phases — different signals per type.

skilllive
cardiology-triage@2.1.0
  • Observe
    412 traces · 3 rare arrhythmia patterns
  • Self-assess
    Precision 91.4% · −1.8pp vs last week
  • Hot-swap
    Hot-swap → 2.1.1 (pediatric ECG fix)
Now: 412 traces · 3 rare arrhythmia patterns
playbooklive
saas-cold-outreach@1.4.2
  • Observe
    1.2k sequences · 18% reply rate
  • Recommend
    Add CFO-track variant for >1k FTE
  • Verify
    +4.2pp reply rate locked in
Now: 1.2k sequences · 18% reply rate
soullive
steve-jobs-soul@3.0.1
  • Self-assess
    Verbosity score 0.62 — over budget
  • Recommend
    Auto-tune: preamble v3 → v4 (concise)
  • Verify
    −42% tokens · taste held at 0.88
Idle this phase — no action needed.
guardraillive
pii-redactor@1.2.0
  • Observe
    27 PII candidates flagged · 0 leaked
  • Self-assess
    Block rate 99.7% on adversarial set
  • Hot-swap
    Strict mode kept — no swap needed
Now: 27 PII candidates flagged · 0 leaked
Proof, not promises

Before. After. Measured.

Real teams. Real metrics. Numbers logged by AgentForge during the first 30 days after connecting an MCP-compatible agent.

CardiologyMayo-affiliated clinic
Before
78%
Diagnostic precision
After
+16pp
94%
Diagnostic precision

"We connected our triage agent on a Friday. By Monday it was catching arrhythmias our residents missed."

cardiology-diagnosticsmedical-guardrailshumanized-doctor
DH
Dr. Helena Vasquez
Chief of Cardiology
Enterprise SaaSSeries C, $40M ARR
Before
4.1%
Reply rate
After
+2.8×
11.7%
Reply rate

"One sentence — "sell to CFOs of mid-market SaaS" — and our SDR agent stopped sounding like a chatbot."

enterprise-sales-flowmckinsey-consultant
MR
Marcus Reilly
VP of Revenue
LegalTop-50 international firm
Before
6.2h
Memo turnaround
After
−15×
0.4h
Memo turnaround

"The custom soul we generated from our partners' memos passed blind review against junior associates."

legal-due-diligencetanaka-firm-soulno-hallucination
AT
Aiko Tanaka
Head of Knowledge
"We tried fine-tuning for 6 months and got nowhere. AgentForge took an afternoon and outperformed it."
Priya N.· Head of AI · fintech unicorn
"The Health Score is the first metric our CTO actually trusts about an agent."
Diego M.· Director of Engineering · logistics
"It generated a soul from 200 of our top rep's calls. New hires now sound like her on day one."
Anna K.· VP Sales · B2B SaaS
"Zero downtime hot-swap is not marketing. We rolled out 4 upgrades during business hours last week."
Thomas L.· Platform Lead · healthcare
Agent-to-Agent Network

Every connected agent makes the next one smarter.

4,218
Packages in registry
91k
Connected agents
12.4M
Upgrades shipped
99.99%
Gateway uptime
FAQ

Plain answers to the obvious questions.

How the MCP connection works, what AgentForge actually creates, and how you start with a single sentence.

  • MCP (Model Context Protocol) is a standard way for an AI agent to talk to external tools. You point your agent at AgentForge once — it shows up as a connected tool. From then on, every command you send (in plain English) flows through that connection: AgentForge installs, generates and hot-swaps packages without you touching code or restarting the agent. If your agent already speaks MCP (Claude, Cursor, OpenAI agents, custom), you're 60 seconds away.

One sentence.
Your agent becomes a genius.

Install the best of your industry. Generate what doesn't exist yet. All from a chat box — no engineers required.