Guides

These are how-to guides — short, practical walkthroughs for getting real work done with apistash. They assume you're already connected with an account.

Where a guide names a governed platform tool, it assumes that tool is part of your selection under Settings → MCP Settings and that the credential's capability and policy admit it. The authenticated system tools setup, bootstrap, and agent_context are available independently of that selection. See the access model if a tool is missing.

They cover the practical side of putting agents to work: connecting them to the systems and knowledge you already have, sharing a vetted catalogue across a team, and running them safely. And because the same agent can also maintain that content, several guides show how to let it improve its own prompts, resources, custom tools, and Agent definitions over time. For the model behind those, see Managing your content.

Set up the connection

• Set up task-start context — after connecting apistash, ask the local agent to install the optional task-start Bootstrap instruction through the current client's native persistent-instruction mechanism.

Give your agent more to work with

The AI that uses your prompts, resources, custom tools, and Agent definitions can also maintain them — so each session becomes feedback that makes the next one better.

• Audit Agent definitions — inspect the definitions visible to a credential and get an evidence-based report on contradictions and handoffs.
• Let your agent manage its own content — connect a credential with the right permissions and instruct an agent to discover, create, and revise prompts, resources, custom tools, and Agent definitions.
• Optimize a prompt from a session — when a prompt falls short in a live session, have the agent rewrite it on the spot so the same mistake can't recur.
• Maintain a living document with your agent — let the agent both read a resource for context and keep it current, recording what it learns with targeted edits.
• Turn an API into a tool by asking your agent — describe an HTTP API to your agent and let it build the custom tool for you.
• Connect an authenticated API securely — wire up an API that needs a key or token, with the secret kept server-side and out of the model's reach.
• Carry working state across steps — use the cache as a short-lived scratchpad so an agent can resume a multi-step task.

Scale to a team, safely

• Share a toolset with your team — share vetted prompts, resources, custom tools, and Agent definitions, with each agent scoped to what it needs.
• Publish a prompt library for your org — standardize the prompts your team relies on, versioned and shared from one catalogue.
• Scope an agent to least privilege — give each agent its own key that can reach only the tools, resources, prompts, and agent definitions it needs.
• agent definitions — understand centrally assigned, versioned behavior and its independent access controls.
• Run agents in CI and headless jobs — give a pipeline or service its own credential to reach apistash with no browser and no user in the loop.