Works today
Claude, Cursor, Codex, Devin, and any MCP-compatible agent connect today.
AI agents (MCP)
Agents search first, then save.
Claude, Cursor, Codex, Devin, and any MCP-compatible agent connect today.
01
Connect any MCP agent.
If it speaks MCP, it can search your knowledge and save what it settles.
Same map
The agent reads the same Corpus a person reads in the app.
Two on Free
Free includes two agent connections. Reading is never metered.
02
Read first, then write.
They search first, then save what they settle.
Search before save
The agent looks up what you already have before it writes.
No empty session
It does not start knowing nothing and invent a parallel page.
Writes are counted
Reads are never metered. Writes follow your plan.
Before
An agent starts blank and files “Architecture v2” next to the real architecture page.
After
The agent searches, finds the page, and updates it.
- Add a retry limit to the payments webhook.
- search("payments webhook retries")
- Payments webhook design · “The provider resends failed events for 24 hours.”
- A page already covers this. Update it, do not start a new one.
- save_knowledge({ topic_id: "payments-webhook-design", content: "Retry limit: 5. Why: the provider already resends for 24 hours." })
- Updated · Payments webhook design · earlier version kept, dated
03
The tools you get.
Read tools are never metered. Write tools land as structured knowledge.
Read tools
search, fetch, explore, get_knowledge_map, query_dataset. Reads are never metered.
Write tools
save_knowledge, capture_team_fact, define_dataset, write_dataset.
Tables stay rows
define_dataset and write_dataset keep tables as rows you can query and total.
Wears the connecting person's badge
Your agent
- Claude, Cursor, Codex, Devin, any MCP agent
reads first
Cluesora MCP
- Read — search, fetch, explore, get_knowledge_map, query_dataset · never metered
- Write — save_knowledge, capture_team_fact, define_dataset, write_dataset · counted
then writes
The map
- Updates the existing page, never a copy
- Old version kept, dated
04
Add this to Cluesora.
Tell your agent “add this to Cluesora” and it lands as properly structured documentation.
One instruction
You do not paste a dump into a folder and hope someone files it.
Checked on the way in
The write gets the same checks and the same permissions as any other way in.
Usable fast
Small writes are usable within about a minute.
05
Extend, do not fork.
Agents update the existing doc instead of forking a new one.
One doc stays one
The settled answer edits the page that already exists.
Sources stay original
Sources are never rewritten. Every answer traces to its origin.
History remains
The new version takes over. The old one is kept and dated.
- Today
Queue workers retry three times, then park the job.
Updated by an agent after the incident review - Last month
Queue workers retry forever.
Kept, dated, still readable
06
The connecting person’s badge.
An AI agent gets exactly the permissions of the person who connected it.
No extra sight
If the person cannot see a source, the agent cannot see it either.
Writes follow access
What the agent saves carries the permissions of the sources it used.
Revoke any time
Disconnect it under Profile → Connected agents.
07
Same machine, same checks.
An agent write is not a side door. It is one more way into the same map.
Same quality gates
Bad input is held back. Repeated passages count once.
Same disagreements
If the write disagrees with a page you already have, both sides stay visible.
Same people map
The agent can ask who owns a service and get a straight answer.
08
Start with the sample.
Every Free workspace comes with a sample dataset. Connect an agent and search it first.
Sample on Free
Free is three seats, a sample dataset, two agent connections, and reading that is never metered.
Search first
Point the agent at search and fetch before you let it write.
Then one write
Tell it to add this to Cluesora, and check that it updated the existing page.
Questions people ask.
Which agents connect today?
Claude, Cursor, Codex, Devin, and any MCP-compatible agent.
Are reads metered?
No. search, fetch, explore, get_knowledge_map, and query_dataset are never metered.
Will my agent create duplicate docs?
No. It searches first, then updates the existing doc instead of forking a new one.
What can the agent see?
Exactly what the person who connected it can see. No more.
How do I try this without our real content?
Use the sample dataset on Free. Connect an agent and search it first.