codai
EU-hosted gateway offering agent memory, web tools and a Claude-routing LLM endpoint
- Platforms
- API, Web, CLI
- Runs
- Remotely
- Version
- 2.1.0
- Listing updated
- today
Similar MCP servers
About codai
codai is an MCP server for ai models and llm tooling. It runs as a hosted remote server, so there is nothing to install locally.
Behind one endpoint, https://ai.codai.ro/mcp, codai serves persistent memory, a model router and optional groups of tools. Its docs describe it as a gateway: besides MCP it speaks an OpenAI-compatible API at /v1 and an Anthropic-style /v1/messages route, so VS Code Custom Endpoint, GitHub Copilot CLI in bring-your-own-key mode, Cursor, Continue, Cline, Zed and Claude Code can all use the same codai_ key.
By default the MCP tool list returns a core set (memory search, store, list, graph and forget, chat, model listing, account and usage) plus web tools, which are switched off for EU-residency accounts. Further groups are opt-in per connection with ?toolsets= or the X-MCP-Toolsets header, covering things like transcripts, schedules, sessions, cards, lists, prompts, sandbox, vault and a Studio group for creative media. Core definitions take about 9 KB of context versus about 90 KB for everything.
The resolve tools let an agent submit a GitHub issue for a quote and accept the work, which debits the quoted price from the wallet and is refunded if no verified fix results. Those need an API key; OAuth tokens are refused.
Install
Run in a terminal
claude mcp add --transport http codai https://ai.codai.ro/mcpAdd to Settings → Connectors → Add custom connector
https://ai.codai.ro/mcpAdd to ~/.cursor/mcp.json
{
"mcpServers": {
"codai": {
"url": "https://ai.codai.ro/mcp"
}
}
}Add to .vscode/mcp.json
{
"servers": {
"codai": {
"type": "http",
"url": "https://ai.codai.ro/mcp"
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"codai": {
"serverUrl": "https://ai.codai.ro/mcp"
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"codai": {
"type": "http",
"url": "https://ai.codai.ro/mcp"
}
}
}Using several servers? Build one config file with the MCP config generator.
Tools this server provides (16)
Taken from the project documentation; the running server may differ.
- memory_search
- Searches stored memories.
- memory_store
- Stores a memory.
- memory_list
- Lists memories.
- memory_graph
- Returns the memory graph.
- memory_forget
- Removes a memory.
- chat
- Chats through the gateway.
- list_models
- Lists available models.
- account_get
- Returns account details.
- usage_get
- Returns usage.
- web_search
- Searches the web (web toolset).
- web_fetch
- Fetches a web page (web toolset).
- deep_research
- Runs deep research (web toolset).
- resolve_submit
- Submits a GitHub issue and gets a quote.
- resolve_get
- Polls a resolve job.
- resolve_accept
- Starts the fix and debits the quoted price; requires confirm_spend.
- resolve_attestation
- Returns public execution proof.
What it does
- Persistent memory
- Search, store, list, graph and forget memories across sessions.
- Opt-in toolsets
- Add groups such as web, cards, schedules or vault per connection to save context.
- Multi-client gateway
- Same key works over MCP, OpenAI-compatible and Anthropic-native endpoints.
- Model aliases
- Choose codai, codai-fast, codai-deep, codai-agent, codai-max, codai-explorer or codai-vision.
- Resolve workflow
- Submit a GitHub issue for a quote, accept it, and poll or attest the result.
- Stdio bridge
- A local bridge script lets clients that cannot send HTTP headers connect.
Good for
- Give Copilot CLI in BYOK mode persistent memory
- Point Cursor or Continue at an EU-hosted OpenAI-compatible endpoint
- Run web search and deep research from an MCP client
- Keep an agent's context small by loading only needed toolsets
codai FAQ
How do I authenticate?
With a codai_ API key; the bridge reads CODAI_API_KEY or COPILOT_PROVIDER_API_KEY from the environment.
Are the web tools always on?
They are included by default but not listed for EU-residency accounts, where they are off.
How do I enable more tools?
Add ?toolsets=cards,web to the server URL or send the X-MCP-Toolsets header; all lists everything.
Which base URL should I use?
https://ai.codai.ro/v1 for OpenAI-style clients, and the root https://ai.codai.ro for Anthropic-native clients.
How do I add codai to Claude Code?
Run this command in your terminal: claude mcp add --transport http codai https://ai.codai.ro/mcp
What are the best alternatives to codai?
The closest alternatives to codai are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.
Where this comes from
- Description: written from the project's website.
- Install steps: from the official MCP Registry entry.


