The Praxium API

Built for people. Built for agents.

Praxium is a Recovery Protocol OS your team uses in the browser and your agents use over MCP: the same studio knowledge, member context and protocols, under one permission model.

Talk to us about API access →Explore the open directory API

Connect your agents

Your studio, in Claude, ChatGPT or any MCP agent.

Praxium runs an MCP server, the Model Context Protocol that Claude, ChatGPT and most agent frameworks connect through. Once your studio is connected, an agent connects with OAuth and works from what your studio knows.

Pull a member's context

The member's goals and plan, scoped to the studio the agent is connected to.

Search studio knowledge

Your protocols, modality guidance and documents, searched for the question at hand.

Draft a protocol for review

The agent drafts from your own protocol library; staff review the plan decision.

Claim and close work

Pick up a follow-up, a reassessment or a content task, then close it with proof of what was done.

OAuth sign-in, tenant-scoped. An agent sees one studio's records and nothing across the line.
People approve the consequential moves. External writes and destructive actions wait for a human yes.
The same Prax your team knows. Prax works from studio knowledge and member context, and staff review important plan decisions.

Praxium reads the whole studio.

Typed core tools cover the work an operator does every day. A gateway finds and runs every other tool on the tenant connection, so an agent never works from a partial view.

Entities: searchEntities · getEntity · filterEntities · createEntity · updateEntity; Members, protocols, modalities, equipment and every other record your studio keeps.
Documents: searchDocuments; Full-text search across the studio's written knowledge.
Context packs: getContextPack; One call that orients an agent: who the studio is, how it works, what matters for this task.
Goal work: listGoalWork · claimGoalWork · closeGoalWork; Claimable work items, so two agents don't pick up the same job.
Skills: list · load; The studio's own procedures, loaded as instructions an agent follows step by step.
Views: list · open; The same filtered lists your team works from in the browser.

The open directory API

Public studio records, one GET away.

Anyone can read the published recovery-studio directory: the same records people browse on askprax.ai, read-only and versioned.

List published studios

No authentication or API key is required. The endpoint supports cross-origin GET, HEAD, and OPTIONS requests and returns JSON.

curl --get "https://askprax.ai/api/v1/directory" \
  --data-urlencode "city=Austin, TX" \
  --data-urlencode "modality=cold-plunge" \
  --data-urlencode "page_size=10"

Filtering and pagination stay predictable.

Parameter	Type	Meaning
q	string · max 80	Case-insensitive match across public names, cities, and modality slugs.
city	string · max 100	Exact published city label, matched case-insensitively.
modality	slug · max 80	Exact lowercase modality slug, such as cold-plunge.
page	integer · min 1	One-indexed page. Beyond the result set returns an empty data array.
page_size	integer · 1–100	Records per page; defaults to 24.

Versioning

Stable operations carry a major version in the path: /api/v1. Backward-compatible fields may be added within v1; breaking changes require a new major path. Deprecated operations remain available for at least 12 months. Before removal, responses identify the policy with rel="deprecation", then publish RFC 9745 Deprecation and RFC 8594 Sunset dates.

Caching

Cache-Control reflects the active public-corpus mode. Follow it, retain ETags when an intermediary supplies them, and use the response links instead of inventing page URLs.

Rate limit posture

No contractual per-client rate limit is advertised today. Cache repeat reads, avoid bursts, and back off if edge protection returns 429. A future fixed quota will ship in this document and response headers before it becomes part of the API contract.

Success is narrow; failure explains the next move.

Stable public studio shape

Every item contains only the public slug, name, city, modality slugs, and canonical profile URL.

{
  "slug": "example-studio",
  "name": "Example Studio",
  "city": "Austin, TX",
  "modalities": ["cold-plunge"],
  "profile_url": "https://askprax.ai/studios/example-studio"
}

RFC 9457 problem details

API errors use application/problem+json, a stable code, a human-readable detail, and a concrete resolution hint. Internal exceptions and upstream bodies are never reflected.

{
  "type": "https://askprax.ai/developers#invalid_request",
  "title": "Invalid request parameters",
  "status": 400,
  "detail": "One or more query parameters are invalid.",
  "instance": "/api/v1/directory",
  "code": "invalid_request",
  "resolution": "Correct the named parameters and retry the same GET request."
}

invalid_request

HTTP 400

A named query parameter is malformed. Read invalid_params, correct it, and retry.

method_not_allowed

HTTP 405

The resource is read-only. Follow the Allow header and use GET, HEAD, or OPTIONS.

api_route_not_found

HTTP 404

The API path does not exist. Recover through the OpenAPI description or API catalog.

directory_temporarily_unavailable

HTTP 503

The public corpus could not be read safely. Honor Retry-After before retrying.

Public discovery stays separate from the platform.

What this API can read

Only recovery-studio fields already visible on the public directory and studio profile pages. The API reuses the directory's admitted, cached projection, so the website and API cannot silently disagree about which studios are public.

What it cannot reach

No private member, tenant, workspace, operator, lead, protocol, assessment, billing, credential, or internal source data. The open directory has no writes, webhooks, OAuth grants, API keys or MCP tools; those live on the platform API, behind sign-in.

Agent-readable by default.

This site describes itself to machines at standard paths, so an agent can find the API, its contract and its rules without a human in the loop.

/llms.txtSite index for language models
/developers/llms.txtDeveloper resources in Markdown
/openapi.jsonOpenAPI 3.1 description of the directory API
/.well-known/api-catalogRFC 9727 API catalog
/.well-known/ard.jsonAgentic resource discovery manifest
/.well-known/ai-catalog.jsonAI catalog entry
/.well-known/agent-skills/index.jsonIntegrity-hashed agent skills
/api/v1/directoryThe open directory endpoint itself

Put Praxium inside the agents your team already uses.

API and MCP access is set up with each studio. Tell us what you want your agents to do and we'll connect them to the right tools, with the right permissions.

Talk to us about API access →