The Legible Connect Tester is the fastest way to check whether your workspace is truly ready to serve useful context to external AI tools. It does more than confirm the endpoint exists. It verifies that the MCP workflow and the knowledge layer behind it are both usable.
This guide helps customers understand each test step, what a warning means, and what to fix before connecting Claude, Codex, Replit, Lovable, or another MCP-compatible tool.
What The Tester Checks
- initialize — confirms the workspace responds as an MCP server.
- tools/list — confirms the expected Legible Connect tools are exposed.
- list_sources — checks whether the workspace has usable content sources.
- search_context — verifies retrieval can return real context chunks.
- get_document — verifies Legible can return a full cleaned markdown document.
Why This Matters
A connection can be technically valid while still being operationally incomplete. For example, the endpoint may be live while the workspace has no synced content yet, or the tools may be exposed while retrieval still returns empty results for real prompts.
The tester helps customers catch that gap before they paste the connection into an external AI tool and expect useful output immediately.
How To Run It
- Open your site in the Legible dashboard.
- Go to AI Export API.
- Create an API key if you have not already.
- Open the Legible Connect section.
- Run the Legible Connect Tester and review each step before sharing the connection with your team.
How To Read The Result
- Pass means the MCP flow is working and the workspace has enough content to serve useful context immediately.
- Warn means the connection works, but the content layer still needs follow-up, such as syncing sources or adding more coverage.
- Fail means the MCP workflow itself is not healthy enough for rollout yet.
Most Common Warning States
- No sources available yet: the endpoint works, but your workspace does not have enough synced content yet.
- Search returned no chunks: the connection works, but the sample query did not match enough content to return useful context.
- Document fetch skipped: the tester could not find a usable source to validate full markdown retrieval.
Important Detail
The tester validates the site-scoped MCP endpoint with your current dashboard session. Your production AI tools use the dedicated alias plus an API key.
Production endpoint
https://mcp.getlegible.com/{workspace_id}
Authentication
Authorization: Bearer lgbl_...Recommended Rollout Flow
- Create an API key.
- Run the tester.
- Make sure sources are visible.
- Make sure search and document retrieval both work.
- Then connect Claude, Codex, Replit, Lovable, or another MCP-compatible tool.