Give every agent exactly what it needs.
Tools to act. Resources to understand. Prompts to stay consistent. Roles that survive the session. All centrally governed.
Works with the MCP clients you already use
- Claude Desktop
- Cursor
- Windsurf
- Cline
- Continue
- OpenCode
- Gemini CLI
- + any MCP-compatible client
Why apistash
One endpoint replaces a stack of integrations
Instead of wiring up a bespoke integration for every tool, dataset, and system, connect them to apistash once — and every agent, on every team, can act on them through a single endpoint.
Act on your systems
Custom HTTP connectors let agents trigger your pipelines, fire webhooks, and call your internal APIs — alongside 70+ built-in tools.
Work from your knowledge
ADRs, coding standards, design guidelines, reports — resources your agents read for the project at hand.
Keep every agent consistent
Shared, versioned prompts so every agent, on every team, follows the same instructions and standards.
Give each agent a durable role
Versioned agent definitions keep identity and behavior consistent, with credential defaults for primary tasks and explicitly delegated roles for subagents.
Your agent, working on your stack
You
Deploy the payments service to staging, then check it against our rollout checklist.
apistash
Agent
Deployed v2.3.1 to staging. Every checklist item passes except the metrics dashboard — flagged for review.
Built-in tools
70+ tools, ready to use
A growing catalogue of 70+ everyday utilities across eight focused categories — ready to add to each agent's governed toolset.
Math
Expression calculator
Time
Datetime, workdays, holidays
Encoding
JWT, TOTP, hashing, Base64/32
Utility
UUID generate & validate
Conversion
Units + JSON/YAML/TOML/XML
Web
URL parse, build, encode
Validation
Email and more
Cache
Store values across sessions
Search, Audio, and Image categories are on the way.
Custom tools
Let your agents act on your systems
Turn any HTTP API into a tool your agent calls by name — trigger a deploy, fire a webhook, hit an internal service. No code, just a declarative connector.
Pipelines, webhooks, any API
Point a connector at your CI/CD, a webhook, or an internal service — method, URL, headers, body, and an input schema.
No code, just a connector
A declarative request definition, not code — edit or swap it any time, with no deploy.
A first-class tool instantly
Once created, your agent calls it by name exactly like a built-in — no wrapper, no glue.
Upstream credentials stay encrypted server-side — never exposed to the model.
// one custom-tool connector
{
"name": "trigger_deploy",
"method": "POST",
"url": "https://ci.acme.com/deploy",
"auth": { "type": "bearer" }
}→ your agent now has a trigger_deploy tool.
Shared knowledge
Give your agents your knowledge, not just tools
Resources are the documents and data your agents read to work on your projects — the decisions, standards, and context they'd otherwise have to guess.
Decisions & standards
ADRs, coding standards, design guidelines — so agents follow how your team actually builds.
Live data & reports
Store reports and generated data over the REST API — from your test pipelines and beyond — and agents always read the latest.
Shared & versioned
One catalogue across your teams; every change is a new version. Agents can even keep it current themselves.
- adr/0007-auth-token-rotation.md
- standards/typescript.md
- design/brand-voice.md
- reports/perf/2026-01.json
Agent definitions
Give every agent a role that survives the session
Define identity and working behavior once, manage it centrally, and use task-start setup to deliver the current active version across clients, projects, and teams.
Assign the right role once
Attach a definition to an API key or OAuth connection. Task-start setup loads it for primary tasks. Parents explicitly give subagents a role, permission to choose one, or no role.
Load depth only when it matters
Core principles arrive at task start. Longer playbooks stay available as conditional context until the work actually calls for them.
Keep every agent on the approved behavior
Every edit creates a clear, immutable history while agents automatically stay on the current active version. If a refinement falls short, bring proven behavior forward again.
Improve the playbook from the work
Agents can list, create, update, and refine agent definitions themselves — within the same permission, ownership, team, and policy boundaries as the other management tools.
Release steward
assigned to staging-deploy-key
Identity
Own safe, evidence-backed releases and make deployment risk visible before acting.
Always on
Verify inputs, protect least privilege, and report evidence with every result.
Delivered by bootstrap
Conditional
Load the rollback playbook when a deployment fails or health checks regress.
Fetched only when relevant
Version recovery
Teams & access
Built for teams — with access you control
One hosted server for your dev and IT teams — internal or external — with fine-grained control over what every agent can reach.
Shared catalogue, scoped per team
apistash catalogue
one shared library
A shared catalogue
One library of prompts, resources, custom tools, and agent definitions — shared, never duplicated.
Per-team toolsets
Scope content per team, so each agent gets exactly what it should.
Scoped credentials
API keys and OAuth, scoped to just what each agent needs.
You decide the reach
Grant access on purpose — a leaked key touches only what you allowed.
Security & privacy
Your data stays yours
apistash is a secure conduit between your agent and the services you connect — nothing more.
Credentials never leave the server
Your upstream API secrets are encrypted at rest and injected server-side — the model only ever sees results.
No training on your data
We never store, analyze, or train on the content of your requests or responses.
Hardened connectors
Custom tools reach public HTTP/HTTPS endpoints only; private, loopback, and cloud-metadata addresses are blocked, redirects are not followed, and responses are capped.
Revoke instantly
Every API key and OAuth client is scoped and can be revoked the moment you need to.
Common questions
Everything you need to know about apistash.
What can apistash do?
apistash is a hosted MCP server that gives your agent four governed primitives through one endpoint: tools — 70+ built-in utilities (math, hashing, JWT & TOTP, encoding, UUIDs, unit & data-format conversion, URL helpers, caching) plus your own custom API connectors; prompts — reusable, versioned templates; resources — documents and data your agent can read; and agent definitions — versioned identities and behavior assigned to a credential. Browse the full catalogue in our documentation. New tools ship regularly.
What is the Model Context Protocol (MCP)?
MCP is an open standard that lets AI assistants (like Claude, and other MCP-compatible agents) securely discover and call external tools, prompts, and resources. Think of it as a universal plugin system for AI agents — apistash implements this protocol so your agent can reach everything you connect through a single server.
How are agent definitions different from prompts?
A prompt is a reusable template for a task you invoke when needed. An agent definition is a durable identity and operating behavior delivered through task-start setup: core instructions load up front, while specialist guidance stays available only when relevant. Assign one to an API key or OAuth connection and enable task-start setup to keep the role consistent across clients and sessions. Every behavior change is versioned, while agents automatically stay on the current active version. Learn how agent definitions work.
Can I add my own API as a tool?
Yes. A custom tool is a declarative connector to any HTTP API. You define the request — method, URL, headers, body — and its input schema in the dashboard, and apistash exposes it to your agent as a first-class MCP tool. Credentials for the upstream API are encrypted at rest and never exposed to the model.
Can my agent create and manage content itself?
Yes. Beyond using them, your agent can create and maintain its own prompts, resources, custom tools, and agent definitions over MCP — adding what's missing and refining what fell short, mid-session. Ask “Set up apistash.” for optional guided starter content: your agent checks existing content and shows drafts and ownership targets for your confirmation before writing. Every management tool follows the credential's permissions and ownership boundaries; Agent assignment, organization-catalogue publication, and connector secrets stay separate dashboard actions. See the guides.
Can my team use apistash together?
Yes. Create an organization to share prompts, resources, custom tools, and agent definitions with your members, and use teams to scope who can reach what. Access is governed by roles and per-credential policies, so each API key or OAuth client only ever sees what you allow.
Do I need an account to use apistash?
Yes — connecting requires a free account and an API key (or an OAuth connection); there is no anonymous access. Creating one is free and takes a minute. You then pick which of the built-in tools your agents reach, and add your own custom tools, prompts, resources, agent definitions, per-key scoping, and organizations & teams.
How do I connect apistash to my AI agent?
Connect to https://api.apistash.io/mcp through your client's native MCP settings. API-key clients send the key in the X-API-Key header; Claude Desktop uses an OAuth custom connector. File names, fields, and setup steps differ by client — follow the client-specific instructions. Grant Tools access to the credential and pick the governed platform tools it should reach under Settings → MCP Settings in the dashboard. Our Get Started guide walks through the connection.
Is apistash free?
Yes — the free tier is genuinely free. Create a free account, connect an API key or OAuth token, and use the built-in tools plus custom tools, prompts, resources, agent definitions, and team features at no cost. Higher-volume Pro and Enterprise tiers are on the way; today everything runs on the free tier, and we will never artificially throttle free usage to push upgrades.
How does apistash handle my API credentials?
Secrets for your custom-tool connectors — the API keys, bearer tokens, or basic-auth credentials of the upstream services you connect — are encrypted at rest and injected server-side when the call is made. They are never returned by the API and never exposed to the AI agent, which only ever receives the structured result.
Does apistash collect or analyze my data?
No. We do not save, analyze, or train on the content of your requests or responses. apistash acts strictly as a secure conduit between your AI agent and the services you connect. The only data we collect is high-level telemetry — tool usage counts and performance metrics — used solely to keep the platform reliable.