anomalyco/opencode
1
0
Fork 0
mirror of https://github.com/anomalyco/opencode.git synced 2026-07-21 17:56:33 -04:00
Code Issues 5.9k Packages Projects Releases 19 Wiki Activity

[FEATURE]: Lazy/dynamic loading for mcp tools #5425

New Issue
Open
opened 2026-02-16 17:52:32 -05:00 by yindo · 0 comments
yindo commented 2026-02-16 17:52:32 -05:00
Owner
Copy Link
Copy Source

Originally created by @DhruvAtreja on GitHub (Jan 13, 2026).

Originally assigned to: @thdxr on GitHub.

Feature hasn't been suggested before.

  • I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Problem

When multiple MCP servers are enabled, all tool definitions load into the agent's context at session startup. A single MCP server like GitHub can add 15-20k+ tokens just for tool schemas. With several MCPs enabled, context bloat can exceed 50k tokens before the user even sends a message. This is even mentioned in the docs: https://opencode.ai/docs/mcp-servers/

Solution

Lazy loading pattern: Instead of loading full tool schemas at startup, we:

Show a lightweight index in the system prompt with server names and tool names only (~150 tokens vs ~1.5k)
Provide a mcp_load_tools meta-tool that loads specific tools on-demand
The agent loop naturally picks up newly loaded tools on the next iteration
Example Flow

System prompt shows:
"playwright (20 tools): browser_snapshot, browser_click, browser_navigate, ..."

User: "Take a screenshot of google.com"

Agent calls: mcp_load_tools("playwright", ["browser_navigate", "browser_snapshot"])
→ Tools loaded into session state

Agent calls: playwright_browser_navigate({ url: "https://google.com/" })
Agent calls: playwright_browser_snapshot()

Related

Originally created by @DhruvAtreja on GitHub (Jan 13, 2026). Originally assigned to: @thdxr on GitHub. ### Feature hasn't been suggested before. - [x] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request Problem When multiple MCP servers are enabled, all tool definitions load into the agent's context at session startup. A single MCP server like GitHub can add 15-20k+ tokens just for tool schemas. With several MCPs enabled, context bloat can exceed 50k tokens before the user even sends a message. This is even mentioned in the docs: https://opencode.ai/docs/mcp-servers/ Solution Lazy loading pattern: Instead of loading full tool schemas at startup, we: Show a lightweight index in the system prompt with server names and tool names only (~150 tokens vs ~1.5k) Provide a mcp_load_tools meta-tool that loads specific tools on-demand The agent loop naturally picks up newly loaded tools on the next iteration Example Flow System prompt shows: "playwright (20 tools): browser_snapshot, browser_click, browser_navigate, ..." User: "Take a screenshot of google.com" Agent calls: mcp_load_tools("playwright", ["browser_navigate", "browser_snapshot"]) → Tools loaded into session state Agent calls: playwright_browser_navigate({ url: "https://google.com/" }) Agent calls: playwright_browser_snapshot() ### Related - Addresses: #514, #1101, #2888, #2953, #3612, #5373, #6024
yindo added the discussion label 2026-02-16 17:52:32 -05:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
acp
bad
bug
discussion
docs
documentation
enhancement
extension
good first issue
growth
help-wanted
mcp
model-problem
needs:compliance
nix
opentui
perf
pull-request
Mirrored from GitHub Pull Request
 tui
web
windows
zen
No Label discussion
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: anomalyco/opencode#5425
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?