Proxy

The proxy is the core of Codecosts. It sits between your Claude client and the Anthropic API, forwarding requests transparently while logging usage.

How it works

Your app → POST /proxy/:slug/v1/messages → Anthropic API
                      ↓
               Usage log created
          (tokens, cost, bodies, timing)

1. You set ANTHROPIC_BASE_URL to your proxy endpoint URL
2. Your Claude client sends requests to the proxy instead of api.anthropic.com
3. Codecosts forwards the request unchanged, including your API key
4. The response is returned to your client — streaming and non-streaming both work
5. Usage is logged asynchronously (tokens, cost, duration, full request/response)

Create a proxy endpoint

Before making proxied calls, create an endpoint in Settings or via the API:

curl -X POST https://codecosts.com/api/proxy-endpoints \
  -H "Authorization: Bearer $CODECOSTS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name": "my-project"}'

Response:

{
  "id": "...",
  "name": "my-project",
  "slug": "my-project",
  "url": "https://codecosts.com/proxy/my-project",
  "created_at": "2026-03-17T00:00:00Z"
}

Send requests through the proxy

POST /proxy/:slug/v1/messages

Same interface as the Anthropic Messages API. Your existing code works unchanged — just set the base URL.

Authentication: Pass your Anthropic API key via x-api-key header or Authorization: Bearer — Codecosts forwards it to Anthropic as-is.

Headers forwarded: All anthropic-* prefixed headers (including anthropic-version and anthropic-beta), x-stainless-* headers (SDK telemetry), x-app headers, user-agent, accept, and accept-language.

Streaming: Fully supported. When stream: true, Codecosts buffers the SSE response, extracts usage from message_start and message_delta events, then returns the full stream.

Rate limit: 300 requests per minute per proxy endpoint (by slug).

Claude Code

export ANTHROPIC_BASE_URL=https://codecosts.com/proxy/my-project
claude "explain quicksort"

Claude Code reads ANTHROPIC_BASE_URL automatically. No other changes needed.

Python

import anthropic

# Option 1: env var (recommended)
# export ANTHROPIC_BASE_URL=https://codecosts.com/proxy/my-project
client = anthropic.Anthropic()

# Option 2: explicit
client = anthropic.Anthropic(
    base_url="https://codecosts.com/proxy/my-project"
)

message = client.messages.create(
    model="claude-haiku-4-5-20251001",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}]
)

TypeScript

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "https://codecosts.com/proxy/my-project",
});

const message = await client.messages.create({
  model: "claude-haiku-4-5-20251001",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello" }],
});

curl

curl https://codecosts.com/proxy/my-project/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "max_tokens": 100,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

POST /proxy/:slug/v1/messages/count_tokens

Passthrough to the Anthropic Count Tokens API. Same authentication and header forwarding as the messages endpoint. No usage logging is performed.

Rate limit: 600 requests per minute per proxy endpoint (by slug).

curl https://codecosts.com/proxy/my-project/v1/messages/count_tokens \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "messages": [{"role": "user", "content": "Hello, how are you?"}]
  }'

What gets logged

Every proxied request to /v1/messages creates a usage log entry with:

Field	Description
model	The model used (from the response, not the request)
input_tokens	Input tokens consumed
output_tokens	Output tokens generated
cache_creation_input_tokens	Tokens used for prompt cache creation
cache_read_input_tokens	Tokens read from prompt cache
cost_cents	Calculated cost based on model pricing
duration_ms	Round-trip time to Anthropic
status_code	HTTP status from Anthropic
request_body	Full JSON request (model, messages, tools, system)
response_body	Full response — JSON for non-streaming, raw SSE for streaming

Failed requests (4xx, 5xx from Anthropic) are also logged with the error body.

Note

The count_tokens endpoint is a passthrough only — no usage log is created.

Proxy endpoints API

List endpoints

GET /api/proxy-endpoints

Returns a paginated list of proxy endpoints for the authenticated user’s organization.

Query parameters:

Parameter	Type	Description
cursor	string	Pagination cursor from previous response
limit	number	Items per page (1–100, default 20)

Response:

{
  "items": [
    {
      "id": "...",
      "name": "my-project",
      "slug": "my-project",
      "url": "https://codecosts.com/proxy/my-project",
      "created_at": "2026-03-17T00:00:00Z"
    }
  ],
  "nextCursor": null,
  "hasMore": false
}

Create endpoint

POST /api/proxy-endpoints

{ "name": "production", "slug": "production" }

slug is optional — if omitted, it’s generated from the name. Must be lowercase alphanumeric with hyphens.

Delete endpoint

DELETE /api/proxy-endpoints/:id

{ "ok": true }