Public plugin/extension distribution for the Within MCP server.
📚 Developer documentation: https://developers.within.ai/
within-mcp packages Within as an installable
plugin for AI assistant clients that speak the Model Context Protocol.
Once installed, the plugin lets your assistant query your organization's processes,
explore the Process Index knowledge graph, and ground answers in how your business
actually runs — not generic guesses.
This repository is the canonical public surface for two install paths:
- Claude Code / Claude — Anthropic's plugin ecosystem.
- ChatGPT / Codex — OpenAI's app ecosystem.
- Gemini CLI — Google's CLI extension ecosystem.
The plugin connects your AI assistant to the Within MCP server at
https://api.within.ai/mcp. Tools the assistant gets access to
include search, fetch, process-hierarchy navigation, evidence retrieval, and
Context Graph traversal. The full skill prompt lives at
skills/within-process-context-graph/SKILL.md.
Customize -> Connectors -> Search -> Within -> Connect
(NOTE: your organization might have to approve this connector, and you should be added as a user into Within for this to work)
Apps -> Search -> Within -> Connect
(NOTE: your organization might have to approve this app, and you should be added as a user into Within for this to work)
From a Claude session, run these two commands one after the other.
1. First, install the marketplace:
/plugin marketplace add Within-AI/within-mcp
2. Next, install the plugin:
/plugin install within@within
Codex splits this into two steps. First, from your shell, register the Within marketplace with Codex:
codex plugin marketplace add Within-AI/within-mcpThen, from a Codex session, run:
/plugins
Search for Within and install the plugin.
The first request that hits a Within tool will prompt you to sign in.
gemini extensions install https://github.com/Within-AI/within-mcpTo login:
/mcp auth within
You will need to be a Within customer to access this app.
The first time the plugin connects, your AI client will prompt you to sign in to Within. the MCP will use whatever authentication is configured by your organization for Within.
Fallback: If your client does not yet support MCP OAuth, you can issue a personal API key from your Within workspace settings and configure your client to send it as a
Bearertoken. Contact hello@within.ai for guidance.
| Path | What it is |
|---|---|
within_mcp/ |
Python package: metadata + builders + CLI for regenerating manifests |
.claude-plugin/plugin.json |
Claude Code plugin manifest (generated) |
.claude-plugin/marketplace.json |
Claude Code marketplace catalog (generated) |
.codex-plugin/plugin.json |
Codex plugin manifest (generated) |
.agents/plugins/marketplace.json |
Codex marketplace catalog (generated) |
.mcp.json |
Shared MCP server config (HTTP transport; used by Claude Code and Codex) |
gemini-extension.json |
Gemini CLI extension manifest (generated) |
skills/within-process-context-graph/SKILL.md |
The Within skill prompt |
tests/test_packaging.py |
Manifest invariants + drift checks |
LICENSE |
Apache-2.0 (covers this shim only — see NOTICE) |
NOTICE |
Trademark + commercial-service notice |
pip install -e ".[dev]"
python -m within_mcp --write # regenerates the public manifests
python -m within_mcp --check # CI-side drift check
pytest -vThe metadata that drives every manifest lives in
within_mcp/metadata.py. Edit there, then re-run
--write and commit the result.
This plugin shim is released under the Apache License 2.0. The
Within service it connects to is a commercial service governed by
Within's Terms of Service. See NOTICE for details.
- Within: https://www.within.ai/
- Privacy Policy: https://www.within.ai/product-privacy-policy
- Terms of Service: https://www.within.ai/terms-of-service-2025
- Issues / contact: mailto:hello@within.ai