gitaiflow MCP — MCP Inspector Verification

It should be used after the native or Docker server has already passed its runtime checks.

#1. What Inspector proves

flowchart LR
    Inspector["MCP Inspector"]
    Init["initialize"]
    Session["MCP session"]
    Tools["tools/list"]
    Schema["tool schemas"]
    Call["tools/call"]
    Result["MCP result"]

    Inspector --> Init --> Session --> Tools --> Schema --> Call --> Result

Inspector verifies the MCP contract.

It does not replace direct gitaiflow testing.

#2. Choose the server

#Native

Start:

python -m mcp_server.server

Connect:

http://localhost:8080/mcp

#Docker

Start the Docker server using:

MCP-DOCKER-VERIFICATION.md

Connect:

http://localhost:8080/mcp

The endpoint is the same; the runtime boundary is different.

#3. Start Inspector

npx -y @modelcontextprotocol/inspector@latest

Connect to:

http://localhost:8080/mcp

Provide:

Authorization: Bearer <MCP_API_TOKEN>

#4. Connection verification

Verify:

server name:     gitaiflow
server version:  1.1.3
transport:       STREAMABLE-HTTP
session:         established
protocol:        negotiated

#5. Tool discovery

Run tools/list.

All six tools must be present:

gitaiflow_change_summary
gitaiflow_last_summary
gitaiflow_changelog
gitaiflow_release_notes
gitaiflow_usage
gitaiflow_list_models

#6. Schema verification

Verify required arguments:

gitaiflow_change_summary.remote
gitaiflow_changelog.remote

Verify optional arguments:

gitaiflow_change_summary.base_branch
gitaiflow_change_summary.path

gitaiflow_changelog.since
gitaiflow_changelog.until

gitaiflow_usage.scope
gitaiflow_usage.since
gitaiflow_usage.until

gitaiflow_list_models.free_only
gitaiflow_list_models.as_json

#7. Tool-call order

A useful first sequence is:

gitaiflow_list_models
        ↓
gitaiflow_usage
        ↓
gitaiflow_last_summary

Then test the tools requiring repository context:

gitaiflow_change_summary
gitaiflow_changelog
gitaiflow_release_notes

Use the actual repository remote. Do not assume origin, gitlab, or another remote name.

#8. Workspace rule for Docker

For the recommended multi-repository Docker layout, /workspace is the allowed parent workspace, not necessarily a Git repository itself.

For example:

/workspace/
├── gitaiflow/.git
├── djangoplay-cli/.git
└── issuetracker/.git

If the tool accepts a repository path, select the repository by name:

path="gitaiflow"

which resolves to:

/workspace/gitaiflow

Git must report that directory as its repository root. You can verify it with:

docker exec gitaiflow-mcp sh -c \
  'git -C /workspace/gitaiflow rev-parse --show-toplevel'

Expected:

/workspace/gitaiflow

Do not change the Docker mount to an individual repository merely to make this call work. The parent workspace is what allows Inspector and Claude to select sibling repositories as well.

#9. Rate limiting

The application default is:

MCP_RATE_LIMIT=5
MCP_RATE_WINDOW_SECONDS=60

Repeated manual Inspector requests can consume this window quickly.

For local development use:

MCP_RATE_LIMIT=100
MCP_RATE_WINDOW_SECONDS=60

After changing .mcp.env, restart the MCP server/container.

A restart invalidates old MCP sessions, so reconnect Inspector.

#10. Successful Inspector gate

Inspector is complete when:

[ ] connection established
[ ] authentication accepted
[ ] initialize succeeds
[ ] session established
[ ] tools/list succeeds
[ ] all six tools visible
[ ] schemas are correct
[ ] required arguments are enforced
[ ] optional arguments are available
[ ] valid tool calls succeed
[ ] MCP results are returned

At this point the MCP protocol boundary is proven.

Next:

MCP-CLAUDE-DESKTOP.md