Integrate with Claude Code / Codex / Cursor

Profy Experts can be consumed as a Skill or MCP Server by mainstream AI coding tools, giving developers access to Expert capabilities without leaving their IDE.

Supported Tools

Tool	Integration Method	Status
Claude Code	Skill / MCP Server	✅ Available
Cursor	Skill	✅ Available
Codex (OpenAI)	Skill	✅ Available
Cline	Skill	✅ Available
Gemini CLI	Skill	✅ Available

Prerequisites

Profy account with Developer role
API Key (create at Developer Console, format sk-pro-*)
Expert must have API access enabled (apiAccess: "open")

Method 1: Skill Integration (Recommended, Zero Infrastructure)

A Skill is a Markdown file that teaches AI coding tools how to call the Profy API. Once installed, the agent automatically knows when and how to invoke Profy Experts.

Claude Code

The plugin ships in the Profy repository under agent-plugin/. Its installer detects which coding tools you have and writes the right file for each of them:

cd agent-plugin && bash install.sh

# Or target specific tools
bash install.sh --only=cursor,codex,claude

It carries four skills — quick-start, distillation, publish-and-earn and evolution — plus a .mcp.json pointing at the remote MCP server. Set your API Key:

export PROFY_API_KEY="sk-pro-your-key-here"

Then simply tell Claude Code:

“Use Profy’s react-architect Expert to review this code”

Claude Code will follow the Skill instructions and call POST https://api.profy.cn/v1/agents/run with expert_identifier: "react-architect" automatically.

Cursor

The same installer writes a .cursor/rules/ entry plus .cursor/mcp.json:

bash install.sh --only=cursor

Codex / Gemini CLI / others

The installer covers 13 tools natively — Claude Code, Copilot, Gemini CLI, OpenCode, Cursor, Codex, Aider, Windsurf, OpenClaw, Qwen Code, Kimi Code, Osaurus and Hermes — each in that tool’s own format and install location. Anything else that speaks MCP can point at https://mcp.profy.cn/mcp directly. Make sure PROFY_API_KEY is exported in whichever shell the tool runs in.

Method 2: MCP Server Integration

MCP (Model Context Protocol) provides deeper integration, allowing AI tools to invoke Profy Experts via structured tool calls.

Configure `.mcp.json`

Create .mcp.json at your project root:

{
  "mcpServers": {
    "profy": {
      "type": "http",
      "url": "https://mcp.profy.cn/mcp",
      "headers": {
        "Authorization": "Bearer ${PROFY_API_KEY}"
      }
    }
  }
}

Register in Claude Code

claude mcp add --transport http profy https://mcp.profy.cn/mcp

Available tools

The MCP server exposes 22 tools, grouped by what they let you do from inside the IDE. Note the emphasis: this surface is built for creating and shipping Experts, not just calling them.

Group	Tools
Identity (2)	`get_creator_profile`, `apply_creator`
Expert config (4)	`create_expert`, `update_expert`, `get_expert_detail`, `list_my_experts`
Skills (3)	`install_skill`, `list_expert_skills`, `list_marketplace_skills`
Billing & publish (3)	`set_billing`, `publish_expert`, `get_submission_status`
Distillation (2)	`start_distillation`, `continue_distillation`
Testing (2)	`test_expert`, `open_expert_chat`
Navigation (1)	`open_profy_page`
Evolution (3)	`list_evolution_suggestions`, `accept_evolution`, `get_evolution_stats`
Earnings & discovery (2)	`get_earnings`, `browse_marketplace`

There is no invoke_expert tool — running an Expert against a payload goes through the REST API (POST /v1/agents/run). MCP covers the authoring loop: distil, configure, test, publish, watch earnings.

API Endpoint Reference

All requests require Authorization: Bearer $PROFY_API_KEY. Base URL: https://api.profy.cn

Method	Path	Description
POST	`/v1/agents/run`	Invoke an Expert once (SSE stream)
POST	`/v1/chat/completions`	OpenAI-compatible chat completions
GET	`/v1/models`	Models available to you
GET	`/v1/meters`	Billing meters (public, no auth)
POST	`/v1/sessions`	Create a session for multi-turn work
GET	`/v1/agents`	List agents you own

The full inventory, including sessions, files and environments, is in the endpoint reference.

Invoke an Expert

The Expert identifier goes in the body, not the path:

curl -N https://api.profy.cn/v1/agents/run \
  -H "Authorization: Bearer $PROFY_API_KEY" \
  -H "X-End-User-Id: your-end-user-id" \
  -H "Content-Type: application/json" \
  -d '{
    "expert_identifier": "react-architect",
    "message": "Review this code for performance issues:\n\nfunction App() { ... }"
  }'

Response: an SSE stream (text/event-stream).

X-End-User-Id is required when calling with an API key. Omitting it returns 400 — it’s what attributes the run to one of your end users rather than to your key itself.

Multi-turn Conversations

Pass session_id to continue an existing thread:

curl -N https://api.profy.cn/v1/agents/run \
  -H "Authorization: Bearer $PROFY_API_KEY" \
  -H "X-End-User-Id: your-end-user-id" \
  -H "Content-Type: application/json" \
  -d '{
    "expert_identifier": "react-architect",
    "message": "How about using virtualization?",
    "session_id": "SESSION_ID_FROM_FIRST_CALL"
  }'

OpenAI-Compatible Chat

curl -s https://api.profy.cn/v1/chat/completions \
  -H "Authorization: Bearer $PROFY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-chat",
    "messages": [{"role": "user", "content": "Hello"}],
    "stream": true
  }'

For Creators: Enable API Access for Your Expert

API access is off by default on every Expert — that’s the stored default of the api_access column, and it carries through releases and pending submissions like any other content field.

Edit your Expert in Studio
Set API Access to Open (any developer holding an API key may invoke it)
Publish a new version — the change reaches the public surface only after review, like every other content change

Once enabled, your Expert can be invoked by developers worldwide from Claude Code, Cursor, and Codex.

FAQ

What's the difference between Skill and MCP?

Skill: Plain text instructions that teach AI how to curl your API. Zero infrastructure, cross-platform compatible.
MCP Server: Structured tool protocol where AI makes function calls. More powerful but requires server-side support.

Start with Skills. Upgrade to MCP Server seamlessly once available.

Getting 403 Forbidden?

The Expert’s api_access field is off, which is the default for every Expert. Ask its creator to enable API access, or use one that already has api_access: open.

How to parse SSE streams?

Responses use standard Server-Sent Events format. Each line starts with data: and the stream ends with data: [DONE]. Most HTTP clients and AI tools natively support SSE parsing.

Which models are supported?

Via the /v1/chat/completions endpoint, all platform-configured models are available (DeepSeek, Qwen, etc.). Available models depend on platform configuration.

How does billing work?

Token-based billing (METERED). Credits are deducted from the API Key owner’s account. View usage in the Developer Console.

Key numbers

Item	Value
MCP tools exposed	22
MCP transport	Streamable HTTP at `https://mcp.profy.cn/mcp`
API key prefix	`sk-pro-`
Default `api_access` on a new Expert	`off`
`X-End-User-Id` on key-authenticated runs	Required — omitting it returns 400
Expert invocation endpoint	`POST /v1/agents/run` (identifier goes in the body)
Response format	SSE, terminated by `data: [DONE]`
Billing	Token-metered, charged to the API key owner

Failures and fixes

Symptom	HTTP	Cause	Fix
`401 Unauthorized`	401	Missing or malformed key	The key must start with `sk-pro-` and travel as `Authorization: Bearer`
`400` complaining about an end user	400	`X-End-User-Id` omitted on a key-authenticated call	Send a stable per-end-user identifier
`400 Missing required field: expert_identifier`	400	Identifier put in the path instead of the body	`/v1/agents/run` takes it in the JSON body
`403 Forbidden` on a specific Expert	403	That Expert’s `api_access` is `off`	Ask its creator to enable it
`404` on `/v1/experts/...`	404	That route doesn’t exist	Use `/v1/agents/run`; see the endpoint reference
The stream ends with no content	—	SSE consumed as a normal JSON response	Read it as `text/event-stream` until `data: [DONE]`
Credits drain faster than expected	—	Token metering bills the key owner, not your end users	Watch usage in the console and cap per end user yourself

Next Steps

Endpoint reference

Every /v1/* route with auth and purpose

SSE events

The event types you’ll parse out of a run

MCP tools

Full parameter reference for all 22 tools

Python SDK

A typed client instead of raw curl

Verified 2026-08-12. Sources: services/core/src/routes/platform-api/index.ts (the /v1 surface), services/core/src/routes/mcp/tools/index.ts (22 registered tools), packages/db/src/schema/marketplace.ts (api_access default).

Get Started

Authentication & Keys

API Guides

SDK

Integrations

Cookbook

Reference

Integrate with Claude Code / Codex / Cursor

Supported Tools

Prerequisites

Method 1: Skill Integration (Recommended, Zero Infrastructure)

Claude Code

Cursor

Codex / Gemini CLI / others

Method 2: MCP Server Integration

Configure `.mcp.json`

Register in Claude Code

Available tools

API Endpoint Reference

Invoke an Expert

Multi-turn Conversations

OpenAI-Compatible Chat

For Creators: Enable API Access for Your Expert

FAQ

Key numbers

Failures and fixes

Next Steps

Endpoint reference

SSE events

MCP tools

Python SDK

​Supported Tools

​Prerequisites

​Method 1: Skill Integration (Recommended, Zero Infrastructure)

​Claude Code

​Cursor

​Codex / Gemini CLI / others

​Method 2: MCP Server Integration

​Configure .mcp.json

​Register in Claude Code

​Available tools

​API Endpoint Reference

​Invoke an Expert

​Multi-turn Conversations

​OpenAI-Compatible Chat

​For Creators: Enable API Access for Your Expert

​FAQ

​Key numbers

​Failures and fixes

​Next Steps

Endpoint reference

SSE events

MCP tools

Python SDK

Supported Tools

Prerequisites

Method 1: Skill Integration (Recommended, Zero Infrastructure)

Claude Code

Cursor

Codex / Gemini CLI / others

Method 2: MCP Server Integration

Configure `.mcp.json`

Register in Claude Code

Available tools

API Endpoint Reference

Invoke an Expert

Multi-turn Conversations

OpenAI-Compatible Chat

For Creators: Enable API Access for Your Expert

FAQ

Key numbers

Failures and fixes

Next Steps