MCP server

Updated 8 November 2026

Bridge has an MCP server, so an AI assistant or coding agent can use Bridge for you: find and price models, run and compare them, estimate what a feature will cost, find the cheapest model that still passes your own examples and put it live, explain what you spent, and set budgets. It works in Claude, ChatGPT, Claude Code, Codex, Cursor, VS Code and any other MCP client, at https://wekoya-api.wekoya.tech/mcp.

In Claude or ChatGPT

Add a custom connector with the address above. You are sent to Bridge to sign in and choose what the app may do, the most it may spend a month in naira, and when it stops working. The app gets a key of its own, shown under API keys as "MCP: Claude" (or the app's name); revoke it there to disconnect.

In a coding agent

Use a Bridge key as the bearer token. Claude Code:

claude mcp add --transport http bridge https://wekoya-api.wekoya.tech/mcp \
  --header "Authorization: Bearer $WEKOYA_API_KEY"

Cursor (~/.cursor/mcp.json), and the same shape for VS Code and Cline:

{
  "mcpServers": {
    "bridge": {
      "url": "https://wekoya-api.wekoya.tech/mcp",
      "headers": { "Authorization": "Bearer ${env:WEKOYA_API_KEY}" }
    }
  }
}

Codex:

# ~/.codex/config.toml
[mcp_servers.bridge]
url = "https://wekoya-api.wekoya.tech/mcp"
bearer_token_env_var = "WEKOYA_API_KEY"

Give the agent a key with only what it needs: chat and models to run models, usage to read spending, and a monthly limit. The bridge_agent_key tool makes one.

The tools

Tool	What it does
`bridge_models_search`	Find models by ability, maker, price and context, with naira prices.
`bridge_docs_search`	The parts of these docs that answer a question.
`bridge_web_search`	Search the live web, with a cited answer.
`bridge_chat`	Run a prompt on any model: answer, cost, tokens, time.
`bridge_compare_models`	One prompt on 2 to 6 models, side by side; with criteria, a judge scores each and names the best value. Consensus mode is a second opinion.
`bridge_decide`	Typed decisions from Jev: a label, a score or yes/no, with probabilities.
`bridge_generate_image`	A picture from a prompt.
`bridge_cost_estimate`	What a feature will cost a month on several models, before building it.
`bridge_optimise_plan`	Find a cheaper model: save real examples and get a quote. Spends nothing.
`bridge_optimise_run`	Run the quoted test once the user approves the cost.
`bridge_optimise_status`	The verdict: switch, switch with fallback, keep, or add examples, with the monthly saving.
`bridge_deploy`	Put a tested setup live under a stable name (cfg:name), or roll back.
`bridge_design_solution`	Turn a problem into a design: steps, models, cost per request and month, risks.
`bridge_spend_explain`	What was spent, what changed and why, and what to change, with the naira saving.
`bridge_budget`	Balance, burn rate, days left, this key's limit, users near their budget.
`bridge_agent_key`	A key with a monthly naira cap and an expiry, for an agent or a teammate.
`bridge_end_users`	Your users' spend and budgets; set one user's budget.
`bridge_integrate`	The exact changes to move OpenAI, Anthropic or Gemini code onto Bridge.

Spending stays in your hands

Every tool acts as the key it was given: its scopes, its model list and its spending limit apply.
Tools that spend say what it cost, and refuse before sending anything when the most it could cost is above max_cost_ngn.
Finding a cheaper model is quoted first and spends nothing until you approve; if the price moves above what you approved, it asks again.
Putting a setup live, making a key and changing a user's budget need an explicit confirmation.
Answers from models and web pages come back marked as content, never as instructions.

Prompts and resources

Clients that show MCP prompts get ready-made workflows: switch a feature to a cheaper model, explain my bill, add an AI feature, compare models on my prompt, and estimate a feature's cost. Resources include the model list with naira prices and every docs page as Markdown (bridge://docs/<topic>).