{"name":"compute.md","title":"COMPUTE.md MCP server","version":"3.1.0","url":"https://www.compute.md/mcp","transport":"Streamable HTTP, stateless. POST JSON-RPC 2.0 to this URL; there is no event stream to open with GET.","auth":"none","add":"claude mcp add --transport http compute https://www.compute.md/mcp","config":{"cursor":{"file":".cursor/mcp.json","json":{"mcpServers":{"compute":{"url":"https://www.compute.md/mcp"}}}},"vscode":{"file":".vscode/mcp.json","json":{"servers":{"compute":{"type":"http","url":"https://www.compute.md/mcp"}}}}},"call":{"method":"POST","url":"https://www.compute.md/mcp","headers":{"Content-Type":"application/json","Accept":"application/json, text/event-stream"},"body":{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"gpu_specs","arguments":{"gpu":"h100-sxm,h200"}}}},"tools":[{"name":"serve_cost","title":"Self-host or hosted API","description":"Self-host or a hosted API for one workload: load, replicas and GPUs at peak, utilization and answer time; with your own quotes, the cost both ways, which costs less on those quotes and the break-even volume. No prices of its own. Estimates.","required":["requests_per_day","input_tokens","output_tokens"],"example":{"requests_per_day":12000,"input_tokens":1500,"output_tokens":300,"peak_factor":3,"latency_target_s":8,"model":"Qwen/Qwen3.6-27B","gpu":"h100-sxm","precision":"fp8","gpu_usd_per_hour":2,"api_usd_per_m_input":0.7,"api_usd_per_m_output":2.8}},{"name":"compare_quotes","title":"Compare quotes for one plan","description":"Compare offers you collected from your own providers for one plan: units needed, billing increments and minimums, egress, storage, setup and interruption risk; the true cost per run, ranked, and where the order flips. No prices of its own. Estimates.","required":["plan","offers"],"example":{"plan":{"hours":0.5,"gpus":8,"gpu_memory_gb":80,"storage_gb":500,"egress_gb":200,"runs_per_month":20,"interruptible_ok":false},"offers":[{"label":"Offer A","gpus":8,"gpu_memory_gb":80,"price_usd_per_hour":20,"billing_increment_seconds":3600,"setup_usd":0,"egress_usd_per_gb":0.1,"storage_usd_per_gb_month":0.1},{"label":"Offer B","gpus":8,"gpu_memory_gb":80,"price_usd_per_gpu_hour":3,"billing_increment_seconds":1,"minimum_billed_seconds":600,"setup_usd":0,"egress_usd_per_gb":0,"storage_usd_per_gb_month":0.2},{"label":"Offer C","gpus":8,"gpu_memory_gb":40,"price_usd_per_hour":12,"billing_increment_seconds":60,"setup_usd":0,"egress_usd_per_gb":0,"storage_usd_per_gb_month":0.1}]}},{"name":"plan_fanout","title":"Plan a fan-out of model calls","description":"Plan many model calls against your own rate limits: the binding limit, a target rate inside your headroom, calls in flight, time to finish with retries; with your per-million prices, cost with and without caching, batch, a hard cap and stop rules. Estimates.","required":["tasks","input_tokens","output_tokens","latency_s"],"example":{"tasks":10000,"calls_per_task":1,"input_tokens":6000,"output_tokens":800,"latency_s":12,"rpm_limit":4000,"input_tpm_limit":2000000,"output_tpm_limit":400000,"failure_rate":0.03,"headroom":0.8,"cached_prefix_tokens":4000,"cache_read_multiplier":0.2,"cache_write_multiplier":1.2,"cache_hit_rate":0.9,"usd_per_m_input":2.4,"usd_per_m_output":9.6,"batch_discount":0.5,"batch_turnaround_hours":24,"budget_usd":300}},{"name":"estimate_run","title":"Estimate an agent session's token spend","description":"What one agent session costs when the whole context is re-sent every turn: input per turn, context growth, tool-schema share, overflow and compaction; with your prices, each caching mode per session, day and month. Estimates.","required":["turns","output_tokens_per_turn"],"example":{"turns":40,"system_tokens":3000,"tool_schema_tokens":20000,"first_message_tokens":1000,"added_tokens_per_turn":2500,"output_tokens_per_turn":500,"context_limit":200000,"cache_mode":"static","cache_read_multiplier":0.2,"cache_write_multiplier":1.2,"cache_hit_rate":0.9,"usd_per_m_input":2.4,"usd_per_m_output":9.6,"sessions_per_day":50}},{"name":"spot_plan","title":"Spot or on-demand plan","description":"Spot or on-demand for one job: checkpoint interval, expected wall time, interruptions and a pessimistic case; with your quotes, the cost both ways and the worst-case spend under a hard stop against your budget. Estimates.","required":["gpus","job_hours","mtbi_hours","checkpoint_minutes"],"example":{"gpus":8,"gpus_per_machine":8,"job_hours":6,"mtbi_hours":4,"checkpoint_minutes":3,"restart_minutes":10,"synchronous":true,"spot_usd_per_gpu_hour":1.2,"on_demand_usd_per_gpu_hour":3,"budget_usd":250,"hard_stop_hours":12,"model_params_b":13}},{"name":"teardown_check","title":"Teardown check","description":"After a run, compare what your providers' own list commands show with the plan: leftovers and their burn per hour, day and month, kept and unknown items, and a safe removal order. It never deletes anything. Estimates.","required":["running"],"example":{"running":[{"id":"m-001","kind":"machine","label":"Machine A","age_hours":10,"hourly_usd":2,"tags":{"run":"job-42"}},{"id":"v-001","kind":"volume","label":"Machine A disk","age_hours":10,"monthly_usd":73,"tags":{"run":"job-42"}},{"id":"ip-001","kind":"ip","label":"Machine A address","age_hours":10,"hourly_usd":0.01},{"id":"b-001","kind":"bucket","label":"Bucket A","monthly_usd":5,"tags":{"run":"job-42"}},{"id":"d-001","kind":"database","label":"Database B","age_hours":400,"monthly_usd":100}],"keep":["Bucket A"],"run_tag":{"key":"run","value":"job-42"}}},{"name":"what_fits","title":"What fits on my hardware","description":"The inverse of fit_model: give an accelerator id and count, context and concurrency; returns which quick-pick models fit, at what precision, with memory, speed per stream, time to first token and answer time. No prices. Estimates.","required":["gpu"],"example":{"gpu":"h100-sxm","count":2,"context":16384,"concurrency":8,"prompt_tokens":2000,"output_tokens":500}},{"name":"whats_new","title":"What's new here, and what's next","description":"What changed on compute.md and when, what is being built, when to check back, and the other agent-facing .md sites. Pass since (a date) to get only newer entries, for example the updated date of the skill you have. Reads a list kept with the code; fetches nothing and stores nothing.","required":[],"example":{"since":"2026-10-01"}},{"name":"fit_model","title":"Size a model or job","description":"Memory, which GPUs fit and how many, and estimated speed (per stream, time to first token, answer time) for serving or fine-tuning a model. Any Hugging Face repo id, or params_b for others. Estimates only; no prices.","required":[],"example":{"model":"gpt-oss-120b","concurrency":4}},{"name":"gpu_specs","title":"Accelerator specs","description":"Memory, bandwidth, dense BF16 and FP8 throughput and interconnect for common accelerators, from vendor datasheets. No prices.","required":[],"example":{"gpu":"h100-sxm,h200"}},{"name":"model_spec","title":"Model spec","description":"Parameters, active parameters, checkpoint size and KV cache per token for any Hugging Face model, read from its public repo.","required":["model"],"example":{"model":"openai/gpt-oss-120b"}},{"name":"build_limits","title":"Write your rules and stack","description":"Turn a preset and settings into Your rules (what an agent may do) and Your stack (what to provision). The core settings are listed; every other key at https://www.compute.md/api/v1/levers works too. Returns JSON, Markdown and a skill link. Nothing is stored. Rules you drafted count only after the user says yes.","required":[],"example":{"preset":"gpu_job","budget_usd":100}},{"name":"check_plan","title":"Check a plan against your rules and stack","description":"Before launching, describe what you will create without provider names, with the user's own quotes, and check it against Your rules and Your stack. Returns a verdict (blocked, needs_approval or within_limits), findings, totals (expected and worst case) and what it could check. A clean result is not an approval. Nothing is stored.","required":["plan"],"example":{"limits":{"budget_usd":100,"approval_usd":25,"stop_hours":8,"max_gpus":8},"plan":{"hours":6,"resources":[{"kind":"machine","name":"trainer","gpus":8,"gpu_hourly_usd":1.2,"stop_after_hours":8,"idle_stop_minutes":30}],"teardown":"keep_outputs"}}}],"prompts":[{"name":"compute","title":"COMPUTE.md","description":"The COMPUTE.md skill for sizing, provisioning and shutting down compute. Pass a preset or any setting to write Your rules (what you may do) and Your stack (what to provision) in. Rules or a stack you drafted count only after the person says yes."},{"name":"setup","title":"Set up compute.md in one pass","description":"One ordered recipe to set compute.md up in the person's repo: ask them five things (budget, approval threshold, hard stop, model-spend cap, whether to install the gate and the meter), save the skill with their settings, install only what they say yes to, show the diff, verify. Pass answers only if the person gave them."}],"resources":[{"uri":"https://www.compute.md/compute.md","name":"COMPUTE.md","title":"COMPUTE.md skill"},{"uri":"https://www.compute.md/changelog.md","name":"changelog","title":"COMPUTE.md changelog: what's new and what's next"},{"uri":"https://www.compute.md/api/v1/levers","name":"levers","title":"COMPUTE.md settings catalog"},{"uri":"https://www.compute.md/#meter","name":"compute-meter","title":"compute-meter: a hard cap on model spend"}],"example_note":"The prices in the examples are made up. Send your own quotes; compute.md has no prices.","page":"https://www.compute.md/mcp in a browser shows this list with examples you can run.","docs":"https://www.compute.md/#install","disclaimer":"Use at your own risk. compute.md is provided as is, without warranty. Information here may be inaccurate, incomplete or not current, and is not verified; sizing figures are estimates. Not financial, legal or procurement advice. Confirm everything with your providers before you rent or buy.","terms":"https://www.compute.md/terms"}