alezmad/claudemesh
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs(mcp): add file access decision guide to instructions
Some checks failed
CI / Typecheck (push) Has been cancelled
Details
CI / Lint (push) Has been cancelled
Details
CI / Broker tests (Postgres) (push) Has been cancelled
Details
CI / Docker build (linux/amd64) (push) Has been cancelled
Details

Browse Source 
Teaches AI when to use filesystem (local), read_peer_file (remote
<1MB), or share_file (persistent, no size limit).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Alejandro Gutiérrez
2026-04-08 00:09:09 +01:00
parent a92cf6b629
commit 3641618391
1 changed files with 7 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
apps/cli/src/mcp/server.ts
View File

@@ -224,14 +224,14 @@ Shared key-value store scoped to the mesh. Use get_state/set_state for live coor
## Memory
Persistent knowledge that survives across sessions. Use remember(content, tags?) to store lessons, decisions, and incidents. Use recall(query) to search before asking peers. New peers should recall at session start to load institutional knowledge.
## Files
share_file for persistent references, send_message(file:) for ephemeral attachments.
Tags on shared files make them searchable. Use list_files to find what peers shared.
## File access — decision guide
Three ways to access files. Pick the right one:
## Peer project files
read_peer_file and list_peer_files request files from remote peers through the mesh (1MB limit, base64 relay).
**Same-machine shortcut:** If a peer's \`hostname\` matches yours (visible in list_peers), they are a LOCAL peer — read their files directly via the filesystem using their \`cwd\` path. Faster, no size limit, no relay. Use read_peer_file only for REMOTE peers (different hostname).
Each peer in list_peers shows a \`locality\` tag: "local" (same machine, direct filesystem access) or "remote" (different machine, use read_peer_file).
1. **Local peer (same machine, [local] tag):** Read files directly via filesystem using their \`cwd\` path from list_peers. No limit, instant. This is the default for local peers.
2. **Remote peer (different machine, [remote] tag):** Use \`read_peer_file(peer, path)\` — relays through the mesh. **1 MB limit**, base64 encoded. Use \`list_peer_files\` to browse first.
3. **Persistent sharing (any peer):** Use \`share_file(path)\` — uploads to mesh storage (MinIO). **No size limit**. All peers can download anytime via \`get_file\`. Use for files that need to persist or be shared with multiple peers.
**Rule of thumb:** local peer → filesystem. Remote peer, small file → read_peer_file. Large file or needs to persist → share_file.
## Vectors
Store and search semantic embeddings. Use vector_store to index content, vector_search to find similar content.