Open Source · Apache 2.0 v0.7.0

Test your AI agents
against the real world

Simulate APIs, databases, MCP servers, and user personas.
Deterministic. Offline. Open source.

Get Started Free Star on GitHub
$ npm install -g @mimicai/cli
mimic — cfo-agent
$ mimic run
$ mimic seed
$ mimic host
PostgreSQL seeded 12 tables, 4,832 rows
Stripe API http://localhost:4101/stripe/v1
Paddle API http://localhost:4102/paddle
Chargebee API http://localhost:4103/chargebee
GoCardless API http://localhost:4104/gocardless
Recurly API http://localhost:4105/recurly
RevenueCat API http://localhost:4106/revenuecat
LemonSqueezy http://localhost:4107/lemonsqueezy
Zuora API http://localhost:4108/zuora
9 MCP servers http://localhost:4201-4209/mcp
Ready in 2.1s

Open Source

Open source.
No vendor lock-in.

The CLI, adapters, MCP servers, and personas are all Apache 2.0. Fork it. Extend it. Run it anywhere.

  • CLI and all adapters are fully open source
  • 13 adapters shipped, 100+ on the roadmap
  • Pre-built personas for realistic testing
  • Build your own adapters with the SDK
  • Runs fully offline — no API keys required
mimic/ # Apache 2.0 ├─ packages/ │ ├─ cli/ @mimicai/cli │ ├─ core/ engine + types │ ├─ adapter-sdk/ build your own │ ├─ adapters/ │ │ ├─ adapter-postgres/ PostgreSQL │ │ ├─ adapter-stripe/ Stripe mock │ │ ├─ adapter-plaid/ Plaid mock │ │ └─ ... 7 adapters │ └─ blueprints/ personas └─ examples/ ├─ billing-agent/ Stripe + PG ├─ finance-assistant/ Plaid + PG ├─ fintech-multi-db/ PG + Mongo └─ ... 9 examples

The Problem

Testing AI agents is broken

Your agent depends on APIs, databases, and tools. But sandbox testing is slow, inconsistent, and incomplete.

Three APIs, three fake users, zero consistency
One persona. Consistent across every system.
Third-party sandboxes throttle your CI tests
Local mocks. Zero latency. No rate limits.
Most sandboxes only cover part of the API
Full API coverage with realistic test data
Sandbox data changes and tests start failing
Deterministic seeding. Identical every run.
MCP servers have no real environment to test against
Mock MCP servers with realistic tool responses

How It Works

Four commands. Full environment.

Point Mimic at your schema, choose a persona, and generate a fully populated local environment with consistent test data.

1
mimic init

Configure

Declare APIs, databases, and MCP servers. Mimic reads your real schema.

2
mimic run

Generate

LLM generates persona blueprints and expands them into realistic rows. Cached for instant reruns.

3
mimic seed

Seed

Push generated data to your configured databases. Atomic, idempotent, and fast.

4
mimic host

Host

Start API mocks and MCP servers locally. One server per adapter, auto-configured transport.


Flagship Example

CFO Agent — 8 billing platforms, one AI

A LangGraph supervisor agent that queries Stripe, Paddle, Chargebee, GoCardless, RevenueCat, Lemon Squeezy, Zuora, and Recurly simultaneously — plus a PostgreSQL product database — through 9 live MCP servers. Ships with a Next.js chat UI.

Browser Next.js UI :3000 CFO Agent :3003
LangGraph Supervisor
postgres :4201
stripe :4202
paddle :4203
chargebee :4204
gocardless :4205
revenuecat :4206
lemonsqzy :4207
zuora :4208
recurly :4209
9
MCP servers
8
billing platforms
1
persona, 6 months data
UI
Next.js chat included
Quick start
$ cd examples/cfo-agent
$ docker compose up -d
$ mimic run && mimic seed
$ mimic host # starts all 9 MCP servers
$ cd agent && npm start
$ cd ui && npm run dev
Ask it anything
“What’s our MRR right now?”
“Give me the full picture for my investor meeting”
“Are any customers paying for a plan they’re not using?”
“Why is our money down this week?”

Architecture

One persona. Every surface.

The Blueprint Engine generates a persona, then adapters project that data across every system your agent uses. Same seed = identical output.

Persona Blueprint
"Alex, 32, fintech PM, $85K, 3 bank accounts,
active trader, Jira power user, Slack daily"
Blueprint Engine
Generate consistent data across systems
PostgreSQL
DATABASE
Real DB seeded
Plaid
API MOCK
:4101/plaid
Stripe
API MOCK
:4102/stripe
Chargebee
API MOCK
:4103/chargebee
Paddle
API MOCK
:4104/paddle

Adapters

Every API your agent needs

13 adapters shipped — 100+ on the roadmap
Fintech & Payments 9 shipped
SStripe
Plaid
PdPaddle
CbChargebee
GCGoCardless
RcRecurly
RCRevenueCat
LSLemon Squeezy
ZuZuora
Fintech & Payments — Roadmap 15 planned
SqSquare
PPPayPal
AAdyen
WWise
CCoinbase
BBrex
RRamp
MMercury
MvMoov
DwDwolla
PsPaystack
FwFlutterwave
MqMarqeta
RvRevolut
RpRapyd
Communication — Roadmap 11 planned
#Slack
TwTwilio
SGSendGrid
Discord
MS Teams
WhatsApp
Telegram
MgMailgun
PmPostmark
VnVonage
MBMessageBird
CRM — Roadmap 7 planned
SFSalesforce
HSHubSpot
PdPipedrive
ZZoho CRM
ClClose
AtAttio
D3Dynamics 365
Ticketing — Roadmap 8 planned
ZdZendesk
Jira
Linear
IcIntercom
PDPagerDuty
FdFreshdesk
SNServiceNow
ScShortcut
Project Management — Roadmap 8 planned
NNotion
AsAsana
TrTrello
MoMonday
AiAirtable
CuClickUp
TdTodoist
BcBasecamp
Calendar & Scheduling — Roadmap 6 planned
GGoogle Cal
CaCalendly
Cal.com
NyNylas
CrCronofy
AcAcuity
Databases 4 shipped
PgPostgreSQL
MMongoDB
MyMySQL
SqSQLite
Developer & Other — Roadmap 8+ planned
GitHub
GLGitLab
ShShopify
S3AWS S3
A0Auth0
MxMixpanel
QBQuickBooks
Google Drive

MCP Servers

First-class testing for MCP servers

10,000+ MCP servers exist in the wild. Almost none have standardized testing infrastructure. Until now.

  • Drop-in compatible

    Our MCP servers mirror real tools — Stripe, Plaid, Paddle, Chargebee, and more. Same tool names. Same schemas.

  • Swap the URL, not the code

    Replace MIMIC_BASE_URL with real credentials. Your agent code stays exactly the same.

  • Works everywhere

    Claude Code, Cursor, VS Code Copilot, or any MCP-compatible runtime.

claude_desktop_config.json
{
  "mcpServers": {
    "mimic-jira": {
      "command": "npx",
      "args": ["-y", "@mimicai/mcp-jira"],
      "env": {
        "MIMIC_BASE_URL": "http://localhost:4000"
      }
    },
    "mimic-slack": {
      "command": "npx",
      "args": ["-y", "@mimicai/mcp-slack"],
      "env": {
        "MIMIC_BASE_URL": "http://localhost:4000"
      }
    }
  }
}

// Swap for production — zero code changes
// "JIRA_API_TOKEN": "real-token-here"

Features

Built for AI agent developers

Everything you need to test AI agents in realistic environments.

Cross-surface consistency

One persona generates consistent data across databases, APIs, and MCP servers. Same user, matching records everywhere.

Deterministic seeding

Same seed + same persona = identical data every run. No flaky tests.

714K rows/sec

High-performance database seeding with FK-aware ordering and atomic transactions.

Works offline

Pre-built personas ship with the package. No LLM calls, no API keys, no internet needed.

Schema-first

Reads Prisma schemas, SQL DDL, or introspects live databases.

100% open source core

CLI, adapters, MCP servers, and personas are Apache 2.0 licensed.


CI/CD

Test agents in every pipeline

Add three lines to your workflow. Every run produces the same environment.

  • Reproducible

    Same persona + same seed = identical environment in every CI run. No drift between runs.

  • Fast

    Local mocks start in under 2 seconds. No external network calls.

  • No secrets needed

    Pre-built personas work offline. No API keys or service credentials in CI.

.github/workflows/agent-tests.yml
- name: Start Mimic
  run: |
    npx @mimicai/cli run
    npx @mimicai/cli seed
    npx @mimicai/cli host --background

- name: Run agent tests
  run: npm test
  env:
    PLAID_BASE_URL: http://localhost:4000/plaid
    STRIPE_API_BASE: http://localhost:4000/stripe/v1

- name: Cleanup
  run: npx @mimicai/cli clean

Pricing

Free to start. Scale when ready.

The open-source core works without an account. Pro unlocks custom personas and advanced testing.

Community
Free
Open source, forever free
  • Full CLI & local generation
  • PostgreSQL seeding
  • MCP server
  • Pre-built personas
  • Community support
Get Started on GitHub
Pro
$39 /seat/mo
For teams shipping agents
  • Everything in Community
  • All domain personas
  • Unlimited blueprints
  • 25K test runs/mo
  • LLM-powered evaluation
  • CI/CD integration
Enterprise
Custom
Compliance & scale
  • Everything in Pro
  • Unlimited test runs
  • SSO / SAML / RBAC
  • Self-hosted option
  • Custom persona building
  • Dedicated SLA

Start testing agents
the way they actually run

Install in 30 seconds. No account required. Pre-built personas ship free. Open source forever.

Read the Docs Star on GitHub
$ npm install -g @mimicai/cli

Join the Pro Waitlist

Mimic Cloud is coming soon. Be the first to know when it launches.