Skip to setup
Open Dashboard
Connect your agent

Open Dashboard for your agent

Connect your agent.

Agents guess at model names and prices. Open Dashboard gives yours exact IDs and provider-published prices with visible units, conditions and source dates across the package's provider catalogues, covering text, image, video and audio models where sources publish them.

Install from npm

Bring live model and price evidence to your MCP agent.

npx -y open-dashboard-mcp
tool count unavailableNode.js requirementMCP version unavailable ↗ Checking npm releases… Reading published package versions from npm. Verify releases on npm ↗

Quick install

npx -y open-dashboard-mcp

Already know your MCP client? Run this line. The wizard below tailors its settings and tool allowlist.

Checking npm downloads… Reading the latest complete weekly window from npm. Verify on npm ↗

Current source evidence

Read a model’s live rate.

These rows are read now from the MCP adapters. Missing source prices stay unknown. Catalogue listing does not establish inference reachability.

Reading current prices…

The contract behind the connection

typed · read-only evidence

Give your selector evidence, not a bare number.

The current MCP contract exposes freshness-aware model evidence and bounded selections for a consuming decision layer. It does not forward inference requests: your provider or agent still runs the model.

dashboard_capability_state

Returns one evidence-shaped row per available model, with explicit public and private decision states and per-field freshness.

dashboard_resolve_model

Resolves requirements into a schema-validated ranked fallback list; unknown values do not count as a favourable match.

Every cost keeps its origin in measurement_origin: fixture is deterministic evidence, live_provider_read is a provider observation, and unknown means no authoritative source was established. Unknown evidence cannot produce a decidable selection, and a cost from one routed provider never appears beside another.

Designed to feed typed decision layers such as System One models. The MCP supplies evidence and selection; it is not a tested integration with those systems.

Current provider catalogue

Sixteen sources, with provenance.

The live website reads all sixteen MCP provider adapters, including KIE, AkashML and io.net. Provider catalogues, rendered price tables and authenticated listings retain their exact amounts, units, conditions and observation time.

Missing or stale authoritative prices stay unknown. A listing is separate from successful inference and a published rate is separate from a settled generation charge.

Read current rates and source status ↗

Step 01 / 03

Where do you work?

We’ll prepare the right settings for your agent.

Choose your agent
Where does your agent run?
Made for the agent you already use.
Codex compatibility & help

This local stdio setup works with Codex in the desktop app, CLI or IDE. They share MCP settings when running on the same host. Add the generated entry to ~/.codex/config.toml there, then restart the connection or open a new session.

Install Node.js as required by the package where Codex runs. Check node --version and npx --version in that environment. If a command is missing, add Node/npm to that host’s PATH and restart Codex. Choose Native Windows for cmd /c npx; choose WSL when Codex runs inside Linux, with Node and the config file inside WSL.

Web and hosted cloud sessions do not inherit this local setup. Open Dashboard does not currently provide a hosted connector. Use codex mcp list to check the saved entry and /mcp in the local app or CLI to check available tools.

Official Codex MCP guide ↗ · Codex in WSL ↗

Claude compatibility & help

Choose Claude Code for its terminal setup, or Claude Desktop for a local Desktop Chat connection. This package runs locally over stdio, so the agent must be able to start a local MCP server.

Claude Cowork and web connectors use remote servers. Open Dashboard does not currently ship a hosted connector for them. Official Claude compatibility guide ↗

OpenClaw compatibility & help

OpenClaw stores saved MCP servers under mcp.servers in its JSON5 config. With the default home, use ~/.openclaw/openclaw.json on macOS, Linux or WSL, and %USERPROFILE%\.openclaw\openclaw.json on native Windows. If OPENCLAW_CONFIG_PATH is set, use that file instead.

Config check: openclaw@2026.9.4 · 11 September 2026. The generated entry was accepted with openclaw mcp set and reported as transport: "stdio" by openclaw mcp status --json. No live openclaw mcp probe has been run here. Run openclaw mcp probe open-dashboard after saving to test the live connection on your machine.

Official OpenClaw MCP guide ↗ · Config path source ↗ · Environment path source ↗

What do I need before connecting?

Install Node.js as required by the package with npm, and use an agent that supports local MCP servers. Your agent starts the npm package through npx. Public model discovery and public usage tools do not need your provider keys; some provider sources and optional account tools need additional configuration. Check the package documentation for those sources.

Can I add or remove tools later?

Yes. Return to this page, choose your tools again, and replace the Open Dashboard entry in your agent’s settings. Restart the connection so it discovers the new selection. The OPEN_DASHBOARD_TOOLS setting controls which tools the server exposes. Omitting it exposes all available tools.

Can this tell me when free models become paid?

It can report observed free-to-paid changes between archived OpenRouter runs. It does not predict a future price change, send alerts or monitor your personal usage allowance. The explorer shows provider free-plan limits separately from model prices.

All available tools Technical reference

The names below are verified against the package contract. Your configuration can enable any subset.

dashboard_catalogue
Model IDs, capabilities, token and media prices.
dashboard_model_economics
Cost and capability ranking, with unknown prices kept visible.
dashboard_resolve_model
Exact model IDs and fallbacks for a set of requirements.
dashboard_capability_state
Freshness-aware candidate rows and public/private decision states.
dashboard_free_models
Models with confirmed zero prices and capability filters.
dashboard_price_comparison
Crazyrouter aliases compared with OpenRouter and sourced reference prices.
dashboard_model_status
Last observed availability for a model ID.
dashboard_whats_changed
Observed model and price changes between archive runs.
dashboard_usage_leaders
Public model and app usage with explicit time windows.
dashboard_matrix
The published app-by-model usage matrix, when available.
dashboard_github_trending
The GitHub trending board and repository links.
dashboard_github_movers
Observed momentum in reviewed AI project categories.
dashboard_benchmarks
Published benchmark observations and source evidence.
dashboard_generation_costs
Measured per-generation costs when the provider reports them.
dashboard_speed
Speed evidence, publication state and measurement protocol.
dashboard_source_health
Data freshness and source availability.
dashboard_contract
Package version, schemas and announced removals.
dashboard_key_inventory
Optional, explicitly configured local key inventory.
dashboard_resolve_seat
Optional, advisory seat selection from a configured model-router; reports unreachable rather than guessing.