Simple pricing for teams of any size

MCP (Model Context Protocol) is an open standard for connecting AI assistants to external tools and data sources. It lets AI clients like Cursor, Claude Code, and Claude Desktop access your GitHub repos, databases, monitoring tools, and more through a unified interface.

The source-available Switchboard runs as a local binary for individual developers. Switchboard Hosted adds multi-tenancy, organizations and departments, SSO, role-based access policies (ABAC), encrypted credential vaults, audit logging, a private-data tunnel for resources behind your firewall, and a web dashboard — everything an organization needs to deploy MCP for a whole team with governance and proof of who can touch what.

Any MCP-compatible client works: Cursor, Claude Code, Claude Desktop, Codex, Windsurf, VS Code (with an MCP extension), and others. Switchboard exposes a standard MCP endpoint that any compliant client can connect to — copy one URL and you're done.

Instead of dumping every tool definition into the model's context, Switchboard exposes two meta-tools: search and execute. Your agent searches for the capability it needs and only the matching tools are returned, and execute can run one tool or chain several across integrations in a single server-side script so intermediate results never round-trip through the model. Adding your 40th integration costs zero prompt tokens on unrelated calls.

Yes, without exposing it. Run the Switchboard tunnel agent inside your network and it dials out to Switchboard over an encrypted connection — no inbound ports to open. Agents can then query private Postgres databases and Kubernetes clusters, with read-only access enforced on your side, not ours. Credentials for those resources stay in your environment and are never stored in the cloud.

All integration credentials are encrypted with AES-256-GCM before they're written to the database — never stored in plaintext. OAuth tokens are protected the same way, and you can rotate or revoke them at any time from the dashboard. For data behind your firewall, the private-data tunnel keeps those credentials in your own environment; they're never sent to or stored by Switchboard. Self-host the source-available build and the encryption key stays entirely on your infrastructure.

Switchboard's compaction pipeline strips redundant fields, columnarizes tabular data, and renders markdown before responses reach the LLM — typically reducing token usage by ~90%. A built-in dashboard tracks savings across four categories (tool catalog, response compaction, markdown rendering, and script intermediates) and can show an estimated dollar-cost reduction in real time.

No. Team plans are a flat monthly price per organization, with a member cap per tier — not a per-seat charge. Add teammates up to your plan's limit at no extra cost. When you outgrow the cap, upgrade to the next tier.

One MCP request to your Switchboard endpoint — typically a single tool call your agent makes (or a search/execute meta-tool request). Browsing the dashboard, connecting integrations, and managing members don't count against the limit. Because execute can chain multiple tools server-side in one request, a multi-step workflow often counts as a single call.

The Free plan is free forever for individuals and small teams — no credit card required. Paid tiers can be started at any time from the dashboard and you can downgrade back to Free whenever you like; downgrades take effect at the end of the billing cycle.

Yes. You can change plans at any time from the dashboard. Upgrades take effect immediately. Downgrades apply at the end of your billing cycle.

You'll receive a warning at 80% usage. Once you hit the limit, MCP requests return an error until the next billing cycle or until you upgrade.

Yes, three ways. Import any OpenAPI or GraphQL spec to turn an existing API into MCP tools instantly. Implement the mcp.Integration interface in Go and contribute it upstream or run it privately. Or, on Business and Enterprise plans, deploy custom integrations as Wasm plugins written in any language that compiles to WebAssembly. All three get the same token optimization, policy controls, and audit logging as built-in integrations.