MCP Server

Shared memory
for AI coding agents

Turn corrections and hard-won knowledge into persistent skill files. Teach your agent once, share with the whole team via Git.

Get started View on GitHub
$ go install github.com/IgorTodorovskiIBM/skillweave@latest
The problem

Agents forget everything

You correct the same mistakes, re-explain the same patterns, session after session. Every team member discovers the same gotchas independently.

The solution

Teach once, share forever

skillweave captures learnings in version-controlled SKILL.md files. Next session, every team member's agent already knows what took you hours to teach.

See it in action

# Bob is porting OSS to z/OS and you spot an issue
You: No, you need to use xlclang++ not g++ — and link with zoslib
Bob: [calls skill_note] Noted.
# when ready, push to share with the team
You: push the skill updates
Bob: [calls skill_push] PR created: #42
# next session — any team member, any machine
Bob: [loads skill_zos_porting_cli] Already knows to use xlclang++ and link with zoslib

How it works

1

Register

Point at a GitHub repo. Works even if the SKILL.md doesn't exist yet.

2

Load

Each skill becomes an MCP tool with a short description. Only fully loaded when needed.

3

Learn

Corrections and discoveries are captured instantly via skill_note.

4

Push

An AI tool merges notes into the SKILL.md, commits, and opens a PR.

5

Share

The whole team's agents pick up the updated skill next session.

MCP tools

Each registered skill becomes its own tool, plus these are always available

skill_note

Capture a one-line learning. Merged into the skill at push time.

skill_push

Merge notes, commit, push, and open a PR — one call.

skill_update

Full rewrite of the SKILL.md with structured learnings.

skill_read

Re-read the skill mid-session to refresh context.

skill_list_notes

Review all unmerged notes before pushing.

skill_<name>

One per registered skill. Loads the skill and starts a session.

Get started in 30 seconds

Install, register a skill, add the MCP config. Done.

$ go install github.com/IgorTodorovskiIBM/skillweave@latest
$ export PATH="$PATH:$(go env GOPATH)/bin"
# point at an existing skill (full URL or user/repo --path)
$ skillweave setup https://github.com/user/repo/blob/main/skills/my-skill/SKILL.md
# paste the printed JSON into .mcp.json or ~/.bob/settings.json