API And MCP | Apologist AI

An Agent can also power your own products and integrations, not just its website, over the Apologist Agent API or over Model Context Protocol (MCP).

Purpose

Turn on programmatic access for an Agent, issue and manage its API keys, and set the defaults those requests use.

This guide covers the Console side only: enabling access, keys, and defaults. Request and response formats, code samples, and endpoint details live in the Agent API documentation.

Prerequisites

An Agent, configured and answering the way you want. Programmatic access exposes the same Agent, so its instructions, sources, and model all apply.
A plan that includes API access.
Somewhere to store an API key securely.

Two Capabilities, One Set Of Keys

API and MCP are separate capabilities that answer different questions:

Capability	What it is for	Description in Console
API	Your own code calling the Agent	”Allow programmatic access to this Agent through an OpenAI-compatible completions endpoint”
MCP	Other AI agents and assistants calling this Agent as a tool	”Allow other Agents to access this Agent’s responses through Model Context Protocol”

They share one thing: API Keys. The API Keys tab appears when either capability is on, and a key issued there works for both. What differs is the endpoint the caller uses.

The API page, with its Defaults, Observability, Semantic Search, and Evaluators tabs, appears only for the API capability. Turning on MCP alone gives you API Keys but no API page, because those defaults govern completions requests rather than MCP tool calls.

Steps

Turn On the Capability You Need

Select Agents in the left-hand navigation, open your Agent, then choose General and select the Capabilities tab.

Under Interfaces, turn on API, MCP, or both. Save, and let the page reload.

An API Keys tab appears at the bottom of the page. If you turned on API, an API entry also appears in the row of pages across the top.

Create an API Key

Select the API Keys tab at the bottom of the page, then choose Create API Key.

The API Keys tab for the Seeker Agent in Console, listing one key named Website integration with an ACTIVE status, its value masked as a row of x characters with a copy button, and Edit and Revoke actions.

Give it a Name that says where it will be used, for example “Website integration” or “Staging”. The name is the only way to tell keys apart later, since the key itself is masked in the table. Leave Active on.

Once created, copy the key from the table and store it somewhere safe.

An API key grants the ability to spend your team’s credits through this Agent. Treat it like a password. Do not commit it to a repository, paste it into client-side JavaScript, or share it in a channel that outlives the conversation. Issue one key per integration so a single compromised key can be revoked without disrupting everything else.

Set the API Defaults

Open the API page from the row of pages across the top, then select the Defaults tab.

The Defaults tab of the Seeker Agent's API page in Console, showing the Default Response Format cards with OpenAI Compatible Stream selected, an unset Default API Key field, and the Strip Markdown from Response Text and Lock System Prompt toggles both off.

Default Response Format decides what a request gets back when it does not ask for something specific:

Format	Use it for
OpenAI Compatible Stream	Streaming, as a direct swap-in for an existing OpenAI integration, using any OpenAI SDK. This is the default.
Plain Text / Markdown	Processing the text yourself, or displaying it somewhere that is not a web page
HTML	Putting the answer straight onto a web page, with Markdown already converted
JSON	Non-streaming calls and platform integrations that need a structured response

Default API Key selects which of the Agent’s active keys this interface uses for its own calls. It stays unset even after you create a key, so choose one deliberately.

Two further options change the response itself:

Strip Markdown from Response Text removes Markdown characters from plain text responses. Turn it on for voice integrations and anywhere Markdown would be read aloud or displayed literally.
Lock System Prompt stops API requests from overriding the Agent’s system prompt. Turn it on when an integration insists on sending its own system prompt and you need the Agent’s configured behavior to win.

Lock System Prompt is the setting that keeps a caller from quietly undoing your work in Customizing Instructions. If an Agent behaves correctly in the browser but not through the API, check this first.

Send Traces To Your Collector, If You Need It

Select the Observability tab. It is editable on Pro and Enterprise (and trial teams, which use Enterprise options). Other plans still get traceparent / traceresponse and notification trace ids; they cannot set a per-agent collector.

Enable OpenTelemetry Export turns the feature on. The endpoint and headers stay disabled until this is on, and a stored URL does not export until you enable it.

OpenTelemetry Endpoint is an HTTPS collector URL, for example https://otlp.datadoghq.com or a Grafana Cloud OTLP gateway. A base URL gets /v1/traces appended. Required when export is on. Private and metadata hosts are rejected.

OpenTelemetry Headers are optional auth headers. Add one row per header (name and value), for example name dd-api-key and your key as the value. Put credentials here, not in the URL. They are stored encrypted as JSON.

Only this Agent’s spans are sent to that collector. Prompt text is never recorded on spans. See Tracing.

Enable Semantic Search, If You Need It

Select the Semantic Search tab. It appears only when your plan includes semantic search.

Enable Semantic Search exposes search over the Agent’s sources through its API, separately from asking the Agent questions. It returns matching passages rather than a generated answer, which suits site search and any feature that needs the underlying material instead of prose.

Cache Duration sets how long results are cached. Individual requests can override it through headers or URL parameters. The field stays disabled until semantic search is on.

Ranking factors for search appear below when your plan includes ranking. They work the same way as the ones in Curating Sources, but they apply to search results rather than to the passages behind an answer.

Enable the Evaluators API, If You Need It

Select the Evaluators tab. It appears only when your plan includes evaluators.

Enable Evaluators API lets your code run this Agent’s evaluators over content programmatically, rather than only inside Console.

Connect an MCP Client

An Agent with the MCP capability exposes a connector at /api/connectors/mcp on the Agent’s own domain, for example:

https://seeker.apologist.bot/api/connectors/mcp

Clients that use the older server-sent events transport connect to /api/connectors/sse on the same domain instead.

The client sends an API key issued above as a bearer token, and a connection without one is refused. Clients that support OAuth can authorize interactively instead of being handed a key.

Once connected, the Agent appears to the calling assistant as a tool it can invoke to ask this Agent a question and get an answer grounded in your sources.

Save And Test

Select Save Changes, then make one real request with the key you issued. Confirm the response format matches what you set and that the answer reflects the Agent’s configured behavior.

Expected Result

Your integration reaches the Agent with its own API key, receives responses in the format you configured, and is governed by the same instructions and sources as the Agent’s website.

Managing Keys Over Time

There are two ways to stop a key working, and they are not the same.

Action	Where	Effect
Deactivate	Inside Edit, by turning Active off	The key stops working but stays listed, so you can turn it back on
Revoke	A row action beside the key	The key is permanently withdrawn

Deactivating is not a row action, which is the part people look for and do not find: open Edit and turn Active off instead.

Deactivate is the right first move when you suspect a problem but are not certain which integration a key belongs to, because it is reversible. Revoke once you are sure.

Where the Endpoint Documentation Lives

Request bodies, parameters, and response schemas are generated from the API specification and kept in one place, so they cannot drift out of date here.

Agent API Overview for what the API covers.
Getting Started for authentication and a first request.
SDKs for the official client libraries.

Troubleshooting

There is no API Keys tab. Neither the API nor the MCP capability is on, or your plan does not include API access.
The API page is missing but API Keys is there. Only the MCP capability is on. The API page belongs to the API capability.
The Semantic Search or Evaluators tab is missing. Your plan does not include that feature.
Observability fields are disabled. Your plan does not include agent.otel (Pro and Enterprise only). You still get HTTP and webhook correlation ids.
Cache Duration cannot be edited. Turn on Enable Semantic Search first.
Requests are rejected as unauthorized. Confirm the key is active and not revoked, and that it is being sent as a bearer token.
The API returns a different personality than the website. The caller is overriding the system prompt. Turn on Lock System Prompt.
Responses contain Markdown characters where they should not. Turn on Strip Markdown from Response Text, or request a different response format.
An MCP client cannot connect. Confirm the MCP capability is on, that the URL is /api/connectors/mcp on the Agent’s own domain, and that the client is sending an API key. Connections without credentials are refused.

Next Step

Continue to the Agent API Overview for the endpoints themselves, or return to Standalone Web UI if you still need the Agent’s own website configured.