Orcha

Orcha MCP server

Orcha is a unified context layer for AI tools and agents. It stores, indexes, and exposes organizational, team, or individual knowledge. Files, structured databases, and connected sources live in one workspace, and agents reach them through Model Context Protocol (MCP), CLI with virtual file system, or REST API, with permissions, citations, and provenance.

This repository is the public connection guide and registry listing for Orcha's hosted MCP server. The service runs at:

https://app.tryorcha.com/mcp

Transport is Streamable HTTP. The application source code is private; nothing needs to be installed or built from this repository to use the server.

What the server provides

30 tools spanning:

• Usage guides the agent can read for itself (get_usage_guide), covering tool choice, search tuning, citations, writing back, and troubleshooting
• Ranked search over all indexed workspace content (search_context) with fast, balanced, and fresh routing modes, plus a citation on every result
• File and folder reading, creation, update, organization, and deletion, scoped to what the connection is allowed to touch
• Curated memories with usage guidance (list_memories, get_memory), including agent-side curation (create_memory, update_memory) covering files, folders, synced documents, and database views
• Structured databases with typed queries (list_databases, query_database, record CRUD)
• Direct access to connected read-only sources (query_source, fetch_source_document, list_source_documents)
• Filesystem-shaped browsing (browse_context): ls, tree, find, cat, grep, and stat over a virtual tree of the workspace
• ChatGPT's two-tool retrieval contract (search, fetch), which chat, deep research, and company knowledge use outside developer mode

Every connection is isolated to the organization, workspace, permissions, and optional folder scope it was granted.

The server also serves its usage guides as Agent Skills under the draft io.modelcontextprotocol/skills extension: skills/list enumerates them and each file is readable at skill://orcha/ / . Clients without skills support get the same content from the get_usage_guide tool.

Connect with OAuth (recommended)

Add the server URL to any OAuth-capable MCP client. The client discovers Orcha's authorization server automatically and walks you through consent; the grant is scoped to the organization and workspace you approve.

Claude Code:

claude mcp add --transport http orcha "https://app.tryorcha.com/mcp"

claude.ai and Claude Desktop: Settings, Connectors, Add custom connector, then enter https://app.tryorcha.com/mcp.

Codex takes TOML rather than JSON. Add this to ~/.codex/config.toml, or .codex/config.toml in a trusted project:

[mcp_servers.orcha]

url = "https://app.tryorcha.com/mcp" auth = "oauth"

Codex also reads Agent Skills from .agents/skills/ in a repository or ~/.agents/skills/ globally, so Orcha's four usage guides can be installed there. They are available with no installation through get_usage_guide.

Cursor and other OAuth-capable clients: add a remote MCP server with the same URL and follow the sign-in prompt.

Claude Code, ChatGPT, and Codex users can instead install the plugins below, which do the same connection and add the usage guides as skills.

Connect with a bearer token (headless and CI clients)

1. Create an Orcha account and workspace at app.tryorcha.com/signup.
2. Create an API token scoped to that workspace with only the permissions the client needs. Tokens begin with orca_.
3. Configure the client with an Authorization header:

{

"mcpServers": { "orcha": { "url": "https://app.tryorcha.com/mcp", "type": "http", "headers": { "Authorization": "Bearer orca_your_token" } } } }

Example prompts

• "Search my workspace for the current authentication requirements and summarize them with the source file names."
• "Show me the five files most related to our deployment runbook."
• "Check which Orcha memories apply to this task and use the best match as the source of truth."
• "Create a Markdown file named launch-checklist.md from these release requirements."
• "Get the engineering-onboarding memory and identify any conflicting guidance."

Claude Code plugin

Claude Code can install the connection and the usage guides together:

/plugin marketplace add westonhancock/orcha-mcp

/plugin install orcha@orcha

The plugin carries its own server configuration, so it replaces the claude mcp add step above rather than following it. It also installs the four usage guides as skills, so Claude loads the relevant one when a task calls for it. See plugins/orcha for details.

OpenAI plugin (ChatGPT and Codex)

The same plugins/orcha directory is also a portable OpenAI plugin: a root plugin.json, an mcp.json that names the server with OAuth dynamic client registration, the four usage guides as skills, and a setup skill that runs after installation to confirm the connected workspace. Once it is listed in the shared ChatGPT and Codex plugin directory, install it from there. Codex can install it from this repository today:

codex plugin marketplace add westonhancock/orcha-mcp

Then open /plugins, choose the orcha marketplace, and install. The marketplace file is .agents/plugins/marketplace.json.

Both plugins are a convenience for their clients. Every other client gets the same guidance from the server through get_usage_guide.

Registry listing

The server is published in the Official MCP Registry as io.github.westonhancock/orcha. The manifest in server.json mirrors the registry entry. A machine-readable server card is served at https://app.tryorcha.com/.well-known/mcp/server-card.json.

Links

• Product: tryorcha.com/product
• Documentation: tryorcha.com/docs/reference/mcp
• Privacy: tryorcha.com/privacy
• Terms: tryorcha.com/terms
• Support: support@tryorcha.com