Agent platforms¶
There are two ways to give ChatGPT, Claude, Copilot Studio, Bedrock or Vertex AI the corpus. Use the first wherever the platform allows it.
| Remote MCP server | Knowledge export (uploaded files) | |
|---|---|---|
| Platforms | ChatGPT connectors and deep research, Claude.ai and Claude Code, Copilot Studio, any MCP client | ChatGPT GPTs and Projects, Claude Projects, Copilot Studio file knowledge, Bedrock Knowledge Bases, Vertex AI, OpenAI vector stores |
| Access control | the live access list, per user, on every call | none: everyone who can use the agent reads every file |
| Deletions and revocations | take effect on the next call | the copy stays until someone deletes it |
| Review gate, audit, exact SQL | yes | no |
MCP¶
- Local (Claude Code, Claude Desktop):
corpusos mcp -C /path/to/corpus -p me@example.com. The principal is fixed for the whole session. - Remote:
corpusos mcp --transport http. Streamable HTTP with OAuth 2.1 access tokens from your identity provider, or API keys. It publishes Protected Resource Metadata so clients can find where to sign in.
Tools: search_corpus, get_document, get_evidence, list_tables, query_table, compare_revisions and get_lineage. It also has search and fetch in the shape ChatGPT connectors expect.
Knowledge exports¶
corpusos export knowledge --target chatgpt --audience sales-team
corpusos export knowledge -t bedrock -a '*' --max-classification public
A document is exported only if all of these hold:
- it is current, approved and indexed;
- its live access list lets the audience read it;
- it is at or below the classification ceiling;
- it has no risk flags.
Each export includes READ-ME-FIRST.md with what the platform won't enforce, and changes.json with what to add, replace and delete since the previous export.
The full reference covers configuration, client setup and the limits used for each platform.