Model Context ProtocolHosted · no install

Give your agent the model catalog

Railwail runs a hosted MCP server. Connect it once and your coding agent searches the catalog, compares models on real prices, works out what a workload costs in US dollars and quotes the API docs, instead of guessing from training data that is months out of date. With an API key, agents can run models directly over MCP: chat, image, video, speech. Billed like the API.

Connect your agent Create API key MCP reference

Endpoint

https://railwail.com/api/mcp

Tools: 13 read · 9 run
Transport: Streamable HTTP; JSON-RPC 2.0 over POST
Protocol: 2025-06-18; also 2025-03-26
Without key: 30 req/min; per IP address
With key: 300 req/min; per account
Currency: USD; 1 credit = USD 0.01

Connect your client

Pick your client, copy the config, done. Without a key it works immediately; the key variant keeps the key in your environment or the client's secret store wherever the client allows it.

Terminal

claude mcp add --transport http railwail https://railwail.com/api/mcp

1Run the command in your project folder. Claude Code keeps the server for this project.
2Add --scope user to make the server available in every project.
3Check it: ask your agent “Which Railwail image models cost the least per image?” It should call search_models.

Official setup docs

Without a key the server only reads catalog, prices and docs. Nothing to sign up for.

Without a key compared with an API key
Aspect	Without key	With API key
Tools	Catalog, prices, rankings, docs, glossary, snippets	All read tools plus chat_completion, run_model, get_job, get_balance
Rate limit	30 requests/min per IP	300 requests/min per account; runs also count against the key's own limit (default 600/min)
Cost	Free	Read tools free; runs billed like the API, in USD
Setup	Paste the endpoint	Create a key, add one Authorization header

Tools your agent gets

This list is what the server returns for tools/list right now. Open a read-only tool and call it: the answer comes from the live catalog, free and without a key. The run tools need your API key, so they are shown here but not called from this page.

search_models

Read-only · free

Search the live Railwail model catalog by free text and filters (category, provider, minimum context window, maximum token price). Returns only models the API can run right now, each with its price and billing unit (per token, per image, per second of output, per 1,000 characters or per GPU-second), context window and latency. Older models carry lifecycle.successor_slug; a query naming a duplicate or removed model finds the model to use instead. Start here for any question of the form 'which model can do X' or 'what does Railwail offer for Y'.

querystring

Free text matched against name, slug, description and tags.(≤ 200 chars)

categorystring

One of: text, image, video, audio, speech_tts, transcription_stt, embedding, code, multimodal, vla_robotics.(≤ 64 chars)

providerstring

Provider slug, e.g. openai, anthropic, google, mistral, deepseek, xai, meta, black-forest-labs.(≤ 64 chars)

tagstring

Require a catalog tag, e.g. vision, coding, svg.(≤ 64 chars)

min_context_windownumber

Minimum context window in tokens.(0–100M)

max_input_price_per_1mnumber

Maximum input price in USD per 1 million tokens. Keeps token-priced models only.(0–1M)

featured_onlyboolean

—

limitnumber

Max results, default 10, cap 50.(1–50)

arguments

copies it as curl

Sent anonymously to /api/mcp. Free; counts toward the 30 requests per minute of your IP.

response

Call the tool to see the live answer from the catalog.

curl

curl -s https://railwail.com/api/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_models","arguments":{"query":"logo","category":"image","limit":3}}}'

What a session looks like

Four typical requests. The tool results below were produced by the real tools from the live catalog when this page was built; press “Run again live” to repeat the exact call.

Which image models can draw a logo with clean lettering, and what does one run cost?

tools/call search_models

{
  "query": "logo",
  "category": "image",
  "limit": 3
}

← resultlive catalog data

{
  "matched": 8,
  "catalog_size": 233,
  "excluded_unavailable": 0,
  "models": [
    {
      "slug": "logoai-replicate",
      "name": "LogoAI (SDXL Logo Generator)",
      "provider": "replicate",
      "category": "image",
      "available": true,
      "price": {
        "unit": "per_gpu_second",
        "credits": 3.72,
        "usd": 0.0372,
        "basis": "Estimate for a typical run of about 31.79 GPU-seconds; the real GPU time is billed.",
        "usd_per_unit": 0.00117,
        "estimated_gpu_seconds": 31.79,
        "hardware": "L40S",
        "default_call_usd": 0.0372,
        "default_call": "Estimate for a typical run of about 31.79 GPU-seconds; the real GPU time is billed."
      },
      "context_window": null,
      "avg_latency_ms": null,
      "url": "https://railwail.com/en/models/logoai-replicate"
    },
    {
      "slug": "ideogram-v2",
      "name": "Ideogram v2",
      "provider": "replicate",
      "category": "image",
      "available": true,
      "price": {
        "unit": "per_image",
        "credits": 9.6,
        "usd": 0.096,
        "basis": "One API call: 1 output at the provider defaults",
        "usd_per_unit": 0.096,
        "default_call_usd": 0.096,
        "default_call": "One API call: 1 output at the provider defaults"
      },
      "fixed_price_usd": 0.096,
      "context_window": null,
      "avg_latency_ms": null,
      "url": "https://railwail.com/en/models/ideogram-v2"
    },
    {
      "slug": "recraft-20b-svg-replicate",
      "name": "Recraft 20B SVG",
      "provider": "replicate",
      "category": "image",
      "available": true,
      "price": {
        "unit": "per_image",
        "credits": 5.28,
        "usd": 0.0528,
        "basis": "One API call: 1 output at the provider defaults",
        "usd_per_unit": 0.0528,
        "default_call_usd": 0.0528,
        "default_call": "One API call: 1 output at the provider defaults"
      },
      "fixed_price_usd": 0.0528,
      "context_window": null,
      "avg_latency_ms": null,
      "url": "https://railwail.com/en/models/recraft-20b-svg-replicate"
    }
  ],
  "catalog_url": "https://railwail.com/en/models"
}

8 image models in the catalog match "logo" and can run right now. The first 3:

• LogoAI (SDXL Logo Generator) (logoai-replicate): ≈ $0.0372 per run (billed on GPU time)

• Ideogram v2 (ideogram-v2): $0.096 per image

• Recraft 20B SVG (recraft-20b-svg-replicate): $0.0528 per image

GPU-time prices are estimates; the run is settled on the real GPU seconds.

Same call, anonymous, free.

Running models with an API key

With a key, agents run models over MCP the same way the API does: same scopes, IP allowlist, rate limit, monthly spending limit and billing. Without a key these tools return an error that explains how to create one.

Tool	Input	Returns	Key scope
`chat_completion` Chat completion (runs a model, needs an API key)	model, messages, max_tokens?, temperature?, response_format?, max_credits?	{ text, finish_reason, usage, credits_charged, max_price_credits, job_id }	chat
`run_model` Run a model (needs an API key)	model, input, max_credits?, wait_seconds? (0–60, default 30)	{ job_id, status: "completed" \| "running" \| "failed", output: { url?, urls?, text?, embedding? }, credits_charged, estimated_credits, max_price_credits, poll_error? }	scope of the model category: images, video, audio, embeddings or chat
`get_job` Get a job (needs an API key)	job_id	same shape as run_model	jobs, or the scope of the job's category
`get_balance` Credit balance (needs an API key)	—	{ credits, usd }	read
`create_fine_tuning_job` Create a fine-tuning job (needs an API key)	model, dataset, dataset_revision?, hyperparameters?, suffix?, max_gpu_hours?, max_credits?	fine_tuning.job { id, status, estimated_credits, held_credits, ... }	fine_tuning
`get_fine_tuning_job` Get a fine-tuning job (needs an API key)	job_id	fine_tuning.job	fine_tuning
`list_fine_tuning_jobs` List fine-tuning jobs (needs an API key)	limit?, after?	{ data: [fine_tuning.job], has_more }	fine_tuning
`cancel_fine_tuning_job` Cancel a fine-tuning job (needs an API key)	job_id	fine_tuning.job	fine_tuning
`list_fine_tuning_events` Fine-tuning job events (needs an API key)	job_id, limit?, after?	{ data: [fine_tuning.job.event], has_more }	fine_tuning

A hard cap per call
max_credits caps the final price: the run is refused before anything is charged when the highest possible price is above it, or when that price cannot be known in advance (GPU-time video and speech models, chat with image, audio or file parts). Useful for agents that loop.
Long runs come back later
run_model waits up to wait_seconds (0–60, default 30). A video that takes longer returns status "running" with a job_id for get_job.
Scopes decide what runs
New keys have read + chat. Tick images, video, audio or embeddings (or all) when you create the key, or those runs are refused.
Same limits as the API
Each key keeps its own rate limit (default 600/min, adjustable 60–6,000), its IP allowlist and your monthly spending limit.
Priced in US dollars
1 credit = USD 0.01. The price is estimated and held before the run, exactly like the API does; failed runs are refunded.
Not everything runs over MCP
Transcription (it needs a file upload), music, 3D and robotics models cannot run over MCP; the tool says so instead of guessing. Speech over MCP returns at most 16 MB of audio, so very long texts are refused before the run; use POST /api/v1/audio/speech for them. Requests to the MCP endpoint are limited to 256 KB.

Questions

Do I need a Railwail account to use the MCP server?

No. Without a key the server answers every read-only tool: catalog, prices, docs and code snippets. You need an account and an API key only when your agent should run models. Without a key the limit is 30 requests per minute per IP address; with a key it is 300 per minute per account.

Can an agent spend money through MCP?

Only with an API key you configured, and only within your balance and monthly spending limit. The run tools use the same rules as the API: the key's scopes, IP allowlist and rate limit, and the same billing. max_credits caps the final price: the run is refused before anything is charged when the highest possible price is above it, or when that price cannot be known in advance (GPU-time video and speech models, chat with image, audio or file parts). Without a key the run tools return an error that explains how to create one.

Where do the prices and rankings come from?

From the same live catalog the website reads, priced in US dollars (1 credit = USD 0.01) with the same formula the API bills. Models that cannot run right now are left out of lists and rankings.

Which transport does the server speak?

Streamable HTTP: JSON-RPC 2.0 over POST with JSON responses, protocol versions 2025-06-18 and 2025-03-26. Any client that speaks Streamable HTTP works; there is no stdio variant, and Claude Desktop reaches it through a custom connector or the mcp-remote bridge.

Endpoints

MCP: https://railwail.com/api/mcp
Discovery: https://railwail.com/.well-known/mcp.json
Crawler summary: https://railwail.com/llms.txt

Building a client or checking details?

Methods, error codes, limits and curl examples are in the MCP reference.

MCP reference Browse models